ACTIVE LABOUR MARKET POLICIES AND LABOUR MARKET TRANSITIONS IN : AN ANALYSIS OF EVENT HISTORY DATA

By AGNE· LAUZADYTµ E·

A dissertation submitted to the Faculty of Social Sciences, University of in partial ful…lment of the requirements of the PhD degree in Economics and Management

Table of Contents

Preface v

Introduction vii

Summary xi

Dansk Resume (Danish Summary) xiii

Chapter 1 1 , Gender and Year Specific Effects of Active Labour Market Programmes in Denmark

Chapter 2 81 Unemployment, Employment and Inactivity in Denmark: an Analysis of Event History Data

Chapter 3 139 Optimal Introduction Time and Length of Active Labour Market Programmes in Denmark

iii

Preface

This thesis was written in 2004 - 2008, during my Ph.D. studies at the School of Economics and Management, University of Aarhus. I’mgrateful to the School of Economics and Management for the excellent research environment and for the …nancial support given to me for participating in courses, seminars, workshops, conferences, and for staying abroad. I received a great help from a number of people while working on this thesis and I would like to take the opportunity to thank them. First of all, I would like to thank my supervisors Michael Rosholm and Peder Pedersen. The completion of this thesis had not been possible without their advice, inspiration and support. I received many valuable comments and suggestions from the Assessment Committee con- sisting of Professor Michael Svarer, School of Economics and Management, University of Aarhus, Senior Research Fellow Knut Røed, Ragnar Frisch Centre for Economic Research, Oslo and Se- nior Researcher Tomi Kyyrä, VATT (Government Institute for Economic Research), Helsinki. I am grateful to the Committee for these comments, which are introduced into the …nal version of the thesis. During the period of November, 2007 - March, 2008 I visited Tinbergen Institute Amster- dam. I would like to thank Jaap Abbring, Gerard van den Berg, Aico van Vuuren and others for the hospitality, interesting seminars and discussions. In addition, thanks to Professor Jan van Ours for his interesting and valuable comments. I would also like to thank Birgitte Højklint Nielsen and Bibiana Paluszewska for the proof reading of the thesis, and the Ph.D. students, the Faculty members and the sta¤ at the School of Economics and Management for the nice and pleasant stay here. And …nally I would like to thank my loving family and friends for understanding and support through these years.

Agne Lauzadyte, Århus, October 2008

v

Introduction

The unemployment rate was high in Denmark during the 1980s and beginning of the 1990s, reaching a peak in the year 1994. In the mid-1990 a comprehensive reform of Active Labour Market Policies

(ALMPs) was enacted. A shift in policy priorities from passive cash support to activation of unem- ployed individuals was implemented by the reform. Since the mid-1990s the use of active labour market programmes has been further increased.

The comprehensive use of ALMPs is often mentioned as one of the explanations of the favourable development in the Danish labour market. It is however a controversial issue whether this was the major reason for the decline in the unemployment rate. Most analyses on Danish data have shown that ALMPs in general have weak or counter-intuitive e¤ects; however, they agree that Private Job training shortens unemployment duration.

The chapters of the thesis County, Gender and Year Speci…c E¤ects of Active Labour Mar- ket Programmes in Denmark and Optimal introduction time and length of Active Labour

Market Programmes in Denmark contribute to the discussion above, and thus are intended as a joint project undertaken to estimate the impact of ALMPs on the exit rates from unemployment of the

UI bene…ts recipients in Denmark.

We use data extracted from an event history data set developed by the Danish National Labour

Market Authority and apply the timing-of-events model introduced by Abbring & van den Berg (2003).

This approach models the process of exit from unemployment into employment and the process of programme entry simultaneously in a multivariate hazard model. We estimate two separate e¤ects of the ALMPs in a duration model: a locking-in e¤ect and a post-programme e¤ect. And …nally, we calculate the net e¤ects of the ALMPs on the unemployment duration.

The reform of ALMPs intended the same rules to be applied in all of the country. The impacts of the policies at the individual level may however di¤er among municipalities for a number of reasons; …rst, the composition of the unemployed workers may di¤er among regions. Second, the labour markets may di¤er, and some labour markets may be more sensitive to ALMPs than others. Third, the type and/or quality of courses o¤ered to unemployed workers may di¤er among counties. County,

Gender and Year Speci…c E¤ects of Active Labour Market Programmes in Denmark studies the regional and annual variations in ALMP impacts, and tries to relate these di¤erences to di¤erences in

vii composition, labour markets, and the characteristics of unemployment and programme use. We create

28 data sets - 14 for men and 14 for women in di¤erent Danish counties (excl. ) - for the analysis, and estimate the impact of each programme separately by counties, and for separate years within each county.

Subsequently, these estimated impacts, by gender, programme, year, and county, are related to a number of explanatory variables, describing the local labour market composition, the composition of programmes, and the structure of local unemployment. This is done by the use of meta-analysis, which is a technique for analyzing and summarizing the results of di¤erent studies, each of which is focused on the same question.

Estimations reveal that only Private Job training reduces unemployment duration, and we …nd a tendency that it performs better in the area around the capital. It is important to stress also that the programme has very favourable e¤ects in Northern , where the unemployment rate is one of the highest in the country.

The e¤ectiveness of ALMPs in relation to men does not seem to depend neither on the composition of the work force nor on the labour market. For women, the sectoral composition does not in‡uence the e¤ect of the ALMPs, but skill composition plays a signi…cant role. The regional unemployment level does not seem to be an important factor. We also …nd a tendency for education programmes and the residual group of other ALMPs to be least e¤ective for the persons with longer unemployment durations.

Besides the main goal of the ALMPs – reintegrating the long term unemployed into the labour market – there is a growing interest in preventive approaches, with activation taking place before a person becomes long term unemployed. The main problem regarding early activation is increasing costs of the programmes, since the target group of participants becomes wider, and there is the risk that the persons, who participate in the ALMPs, would have found a job anyway, i.e. the deadweight costs becomes higher. Optimal introduction time and length of Active Labour Market Programmes in Denmark examines the e¤ects of the programmes, depending on the time spent in unemployment before activation. Another goal of this chapter is to discover how the performance of the two types of

ALMPs - education and Other ALMPs - varies with the length of the programmes.

The activation time dependent analysis leads to a conclusion in favour of programme participation of the unemployed individuals in the second half year of the UI bene…ts spell. Very early activation is not supported by the results, while the analysis of the impact of Education and Other ALMPs depending

viii on the length of the programmes …nds that only short term programmes of up to 6 weeks reduce unemployment duration.

Persons over …fty are found to be the most sensitive group to the in‡uence (both positive and negative) of ALMPs, and Private Job training and the short term activation are found to be highly e¤ective in all stages of their unemployment spell. For the 30-49 years old individuals, however, the programmes are found to be ine¤ective after two years spent on UI bene…ts. Thus, the long-term unemployed in Denmark seem to have speci…c problems, which can’t be addressed by participation in the programmes.

A register based sample of the labour market spells of 20-59 old individuals in a representative 1 per cent sample of the 16-70 year old Danish population in 1994 –2003 leads to the result that nearly half of the employment spells of the persons, who in the previous spell were unemployed, end in new unemployment, while those in the previous spell inactive tend to return back into inactivity. Unem- ployment, Employment and Inactivity in Denmark: an Analysis of Event History Data tackles the problem described above by employing a set of the explanatory variables representing per- sonal and geographical characteristics of the individuals, and their labour market history, and estimating their e¤ect on the labour market transitions.

We use the longitudinal register-based data and estimate a discrete time hazard model for the exits from the di¤erent labour market states – unemployment, employment and inactivity – in the Danish labour market. We distinguish between the di¤erent possible destination states, adopt a competing risks formulation and run multinomial logit estimation.

The estimations results show that women and elderly individuals face a higher risk of leaving the labour market and experience higher survival in both unemployment and inactivity. We also …nd un- skilled and low-educated persons, and residents of the biggest Danish cities (except the individuals over

…fty), to be disadvantaged in the labour market. Being previously employed reduces the risk of leaving the labour force and increases the re-entry to employment probability, while long-term unemployment or inactivity makes workers more likely to get back into these labour market states in the future.

Another important …nding is a break in the transition rates from employment and inactivity after

12 months spent there: those who survived in a job for at least one year tend to remain employed, while persons inactive longer than year face a much higher risk of marginalisation from the labour market.

ix

Summary

This Ph.D. thesis, comprising three self contained chapters, focuses on labour market transitions of the UI bene…ts recipients in Denmark. Even thought the papers are self-contained, the …rst and the third paper could be seen as a joint project undertaken to estimate the impact of Active Labour Market Programmes (ALMPs) on the exit rate from unemployment, while the second chapter is devoted to the transitions in a three-state labour market. The …rst paper in the thesis is focused on regional di¤erences in the e¤ects of programmes, i.e. the timing-of-events model developed by Abbring & van den Berg (2003) is used to estimate the impact of programmes on the escape rate from unemployment in Danish counties. As programme impacts are estimated separately by county and year, a meta-analysis, relating the net impacts of each programme to a number of explanatory variables, is conducted. There is found a tendency for public job training and education programmes to be least e¤ective for the persons with long unemployment durations. The second paper estimates a discrete time hazard model for the exits from the di¤erent labour market states - unemployment, employment and inactivity (or OLF) - in the Danish labour market. The estimations results show that women and elderly individuals, and the less educated and unskilled workers face a higher risk of a transition to a state outside the labour market and experience higher survival in both unemployment and inactivity. Being previously employed reduces the risk of OLF and increases the re-entry to employment probability, while long-term unemployment or inactivity makes workers more likely to get back into these labour market states in the future. The third paper is using the timing-of-events model and examines the e¤ects of the pro- grammes, depending on the time spent in unemployment before the entry into the programme, and on the length of the programmes. It is found that only Private Job training, and short term Education and Other ALMPs reduce unemployment duration, if the activation takes place in the …rst two years of the UI bene…ts spell. Generally, the results favour the activation of the unemployed persons in their …rst year of unemployment. However, activation in the …rst 1-6 months of UI bene…t spell is not supported. ALMPs in Denmark are found to be ine¤ective after two years spent on UI bene…ts.

xi

Dansk Resume (Danish Summary)

Denne afhandling består af tre selvstændige kapitler og fokuserer på arbejdsmarkedsovergange for UI dagpenge modtagere i Danmark. Selvom artiklerne er selvstændige, kan den første og den tredje af dem ses som et samlet projekt med henblik på at estimere ALMPs’ virkning på afgangsfrekvensen fra arbejdsløshed, mens det anden kapitel er dedikeret til overgangene i tre-tilstands arbejdsmarked. Det første papir i afhandlingen er fokuseret på regionale forskelle i e¤ekter af programmer, dvs. benytter timing-af-events modellen udviklet af Abbring & van den Berg (2003) med henblik på at estimere ALMPs’virkning på afgangsfrekvensen fra arbejdsløshed i de Danske amter. Da regionale og årlige programvirkninger er estimeret separat, en metaanalyse, for at relatere nettovirkningerne af hvert program til et antal forklarende variabler bliver udført. Der er fundet en tendens til at den o¤entlige jobtræning og uddannelsesprogrammer er mindst e¤ektive for personerne med lang arbejdsløshed. Det andet papir estimerer en diskret hasard model af afgang fra de forskellige arbejds- markedstilstande - arbejdsløshed, beskæftigelse og inaktivitet (eller uden for arbejdsstyrken, OLF/UFA) - på det danske arbejdsmarked. Resultaterne viser at kvinder og ældre individer, også de lavt uddannede og ufaglærte arbejdere, har større risiko for at møde marginaliserin- gen på arbejdsmarkedet. Tidligere beskæftigelse reducerer risikoen for OLF/UFA og forhøjer sandsynligheden for at …nde job, mens langtidsledighed eller inaktivitet giver arbejdere større sandsynlighed for at vende tilbage til disse arbejdsmarkedstilstande i fremtiden. Det tredje papir benytter timing-af-events modellen og undersøger programmernes e¤ekter, afhængende af arbejdsløshedsvarigheden før indtræden i programmet, og længden af aktiver- ing. Vi …nder at kun private job-træning og kort sigtet uddannelse eller residual kategorien ”andre ALMPs” reducerer arbejdsløshedens varighed, hvis aktiveringen foregår i de første to arbejdsløsheds år. Generelt, skal de arbejdsløse personer aktiveres i det første arbejdsløsheds år. Men resultaterne støtter ikke aktivering i de første 1-6 måneder med modtagelse af UI. Programmerne i Danmark fandt vi ine¤ektive efter to år med UI dagpenge modtagelsen.

xiii

Chapter 1

County, Gender and Year Speci…c E¤ects of Active Labour Market Programmes in Denmark

County, Gender and Year Speci…c E¤ects of Active Labour Market Programmes in Denmark

Michael Rosholm Department of Economics, Aarhus School of Business, University of Aarhus

Agne Lauzadyte School of Economics and Management, University of Aarhus

October 6, 2008

Abstract

We use the timing-of-events model – developed by Abbring & van den Berg (2003) for identifying treatment e¤ects non-parametrically in a duration model framework –to estimate the impact of ALMPs on the escape rate from unemploy- ment in Danish counties. We distinguish between two separate e¤ects - a locking-in e¤ect and a post-programme e¤ect - and estimate the model separately for each county, and for men and women. Subsequently, we calculate the net e¤ect of ALMPs on unemployment duration. We …nd that only private sector employment subsidies reduce unemployment duration. As program impacts are estimated separately by region and year, we subse- quently conduct a meta-analysis, relating the net impacts of each programme to a number of explanatory variables. We …nd a tendency for public job training and education programmes to be least e¤ective for the persons with long unemployment durations. This important result has a number of policy implications, as the implied trade-o¤ between equity and e¢ ciency in the construction of ALMPs would tend to favour other policies than these two, which unfortunately are the programmes which are used most intensively.

Keywords: Active labour market policy, Timing-of-events, Duration model. JEL-code: C41, J64

3 1. Introduction

Unemployment was high in Denmark during the 1980s and 90s, reaching a record level of 12.3% in 1994. Consequently, there was a perceived need for new actions and policies in the combat of unemployment, and a law on Active Labour Market Policies (ALMPs) was enacted in 1994. The instated policy marked a dramatic regime change in the intensity of active labour market policies. After the reform, unemployment has decreased signi…cantly from the peak in 1994 to 4.5% in 2006.

Table 1. Unemployment In Danish Counties (Excl. Bornholm) In 1990-2006, %

1990 1992 1994 1996 1998 2000 2002 2004 2006 Country 9.7 11.3 12.3 8.9 6.6 5.4 5.2 6.4 4.5 and 12.3 14.9 16.0 12.8 8.8 5.7 5.8 6.9 5.6 6.9 9.2 10.6 7.9 5.6 4.2 4.1 5.3 4.1 6.6 8.4 9.7 6.9 4.8 3.7 3.7 4.5 3.1 7.0 8.8 9.7 7.2 4.9 3.8 3.8 4.6 3.2 Western Zelland county 10.9 12.0 13.0 9.3 6.8 5.6 5.2 6.7 4.7 Storstrøms county 11.5 12.8 14.3 10.6 8.3 6.6 6.2 6.6 4.6 11.1 12.7 14.1 8.9 6.7 6.5 6.0 7.3 5.1 Southern Jutland county 9.6 10.6 10.8 7.2 5.4 5.2 5.3 6.4 4.1 county 9.0 9.9 9.9 7.0 5.2 4.6 4.5 5.2 3.1 9.2 10.7 11.3 7.6 6.0 4.8 4.9 6.1 3.6 Ringkøbing county 7.7 8.4 8.8 6.4 4.8 4.1 4.1 5.3 3.4 Århus county 10.5 12.0 12.8 9.3 7.2 6.2 6.0 7.1 4.6 8.6 9.5 9.6 7.2 5.1 4.6 4.3 4.9 3.7 Northern Jutland county 12.9 14.5 15.1 10.7 8.1 7.2 6.8 8.7 6.8

However, the unemployment rates and their evolution over time di¤er among Danish coun- ties, see Table 1. It has fallen by astonishing 64% in the Municipalities of Copenhagen and Frederiksberg, while in Southern Jutland the decline was more modest, slightly above 50%. In two of the traditionally most problematic counties, Northern Jutland and Storstrøm, the unemployment rate fell from a level around 14-15% and by 55% and 57%, respectively. Though the unemployment rates show some variation, the relative changes from business cycle trough to peak is of the same order of magnitude in all municipalities.

4 The goal of the new labour market policies was to upgrade the skills of long-term unemployed to bring them back into employment. The ALMPs are seen as an important part of the Danish ’Flexicurity’ model; see e.g. Andersen & Svarer (2006), which consists of ‡exible hiring and …ring rules in the Danish labour market, a fairly generous unemployment insurance system, and a fairly strict set of rules and regulations regarding availability for work. The aim of ALMPs in Denmark is twofold; …rst, they are designed to upgrade the skills of the unemployed workers to improve their chances of …nding a job, and second, they serve as availability tests, and to some extent as ’threats’(for more on threat e¤ects, see Black et al., 2003, and Rosholm & Svarer, 2004). As a consequence of this dual purpose, there has been some debate over the impacts of the programmes in terms of their most important purpose, the upgrading of skills; indeed, most analyses that have been conducted have shown that ALMPs in general have quite disappointing e¤ects, see e.g. Rosholm & Svarer (2004) and Munch & Skipper (2004). Most analyses on Danish data seem to agree, however, that subsidised employment in the private sector shortens unemployment duration; see Bolvig et al. (2003) and Graversen (2004). The law on ALMPs implied the same rules regarding eligibility and participation in pro- grammes to be applicable in all regions of the country, but it may still be the case that the impacts of these policies at the individual level di¤er among municipalities for a number of rea- sons; …rst, the composition of the unemployed workers may di¤er among regions, and second, the labour markets may di¤er, and some labour markets may be more receptive of ALMPs than others. Third, it may be the case that the composition of instruments used, or the type and/or quality of courses o¤ered to unemployed workers, may di¤er among counties. In this paper we study regional and annual variations in ALMP impacts, and try to relate these di¤erences to di¤erences in composition, labour markets, and the characteristics of un- employment and programme use. We …rst estimate timing-of-events duration models, following local tradition in the programme evaluation literature (see Rosholm and Svarer, 2004). First, the duration model comprises a dynamic approach to the evaluation of ALMPs, and second, the timing of events model of Abbring & van den Berg (2003), allows for selection of observed as well as unobserved variables under relatively mild assumptions. The e¤ects of ALMPs are separated into two e¤ects - a locking-in e¤ect and a post-

5 programme e¤ect. During the program participation period, job search intensity may be lowered since there is less time to search for a job, and because the individual might want to complete an ongoing skill-enhancing activity. Therefore, a decline in the job …nding rate is expected during this period, and this is called the locking-in e¤ect. The post-programme e¤ect covers the period after participation in a programme. If an individual’semployability has been increased by pro- gramme participation, an increase in the job …nding rate is expected. The combination of these two e¤ects consequently determines the net e¤ect of ALMPs on unemployment duration, which is also calculated. The net impact on the expected duration of unemployment is calculated for each of four di¤erent programmes: private sector employment subsidies, public sector tempo- rary jobs, education and training programmes, and a residual group of other programmes. The impact of each programme is estimated separately by counties, and for separate years within each county. Subsequently, these estimated impacts, by gender, programme, year, and county, are related to a number of explanatory variables, describing the local labour market composition, the composition of programmes, and the structure of local unemployment. This is done by the use of meta-analysis, which is a technique for analyzing and summarizing the results of di¤erent studies, each of which is focused on the same question1. The idea to use meta-analysis to measure the e¤ectiveness of European ALMPs was …rst implemented by Kluve and Schmidt (2002), who summarized a total of 53 European active labour market programmes. Further it was developed by Kluve (2006), when he summarized e¤ects of European ALMPs from all European evaluation studies available to that date (95 di¤erent evaluation studies). In Kluve (2006), the implementation of the analysis …rst considers a binomial outcome, i.e. whether the evaluation of a programme …nds a positive treatment e¤ect or not. The framework of probit regression is used, where education programmes are taken as the base category, i.e. the e¤ects of the ALMPs are judged relative to these programmes. Kluve (2006) …nds that

1 Choosing one-step estimation, i.e. including regional factors directly into the hazard model, would gain e¢ ciency in estimations, however, there are several reasons for the choice of two step estimation: - Computational reasons. Currently we estimate on 28 datasets, and this allows us to use 25% of observations in the data sets (though computation time is very long). Choosing to run the datasets for the whole country would require us to reduce the sample to 3-5% of the data sets. Therefore the number of observations would decrease and there would be more statistical uncertainty in the estimations; - Including regional factors directly into the hazard rate model for each county, the local characteristics would not show much variation, since they would then only vary over time within county.

6 education programmes in European countries tend to produce a modest positive impact on the post-programme employment rates. He also reports that private sector incentive programmes and Services and Sanctions show a much better performance - they are 40-50 % more likely to give a positive impact than the traditional education programmes. He concludes with the policy advice, that education programmes should be continued and private sector incentive schemes should be fostered. The structure of the paper is the following: the next section describes the elements of Danish Active Labour Market Policies. The econometric model is explained in section 3, while section 4 presents the data and section 5 - the Meta analysis technique. Estimation results are discussed in section 6, section 7 presents estimates of other characteristics, section 8 - the Meta analysis …ndings, and section 9 concludes.

2. Danish Active Labour Market Policies

The Danish unemployment system is split into two parts, depending on whether the concerned unemployed individual is eligible for unemployment insurance (UI) bene…ts or not. Approxi- mately 20% of the labour force are not eligible for UI bene…ts, and they may instead receive unemployment assistance, which is administrated by the municipalities, and where the rules and regulations concerning active policies di¤er slightly from those that apply to UI recipients. As this paper is mainly concerned with UI recipients, in the following we will present only the policies that apply to individuals receiving UI bene…ts. The "rights and obligations" principle is the key principle of the current Danish labour market policy. The principle is based on the right of individual to the compensation for the loss of income, but also on the obligation to take action to get back into employment. The society has the obligation to help the individual to improve his situation on the one hand, but on the other hand the society has the right to make requirements to the individual concerned. The length of the period during which an unemployed individual can receive UI bene…ts has been reduced signi…cantly since the 1994 reform. Before the reform in 1994, participation in programmes led to renewed eligibility for UI bene…ts. In 1994, this renewal of eligibility rule was abandoned, such that only ordinary employment during a period of time would lead to

7 renewed eligibility. In 1994, the maximum UI bene…t duration was 7 years, including a 4 year ‘passive’period and subsequently a 3 year ‘active’period. This duration has gradually been reduced from 1994 to 1999 such that currently the maximum UI bene…t period is 4 years. After 4 years of UI bene…t receipt, an individual must have at least 26 weeks of full time employment in order to renew bene…t eligibility. Under current rules the passive period lasts 9 months and every unemployed individual, older than 30, who is unemployed for more than 9 months, is required to participate in active labour market programmes. If a person is still unemployed after 26 weeks since programme completion, he is required to participate in another ALMP. Programmes can also be o¤ered during the passive period, based on the regional labour market council’s evaluation of the regional needs, or in order to test the availability for work of a certain individual or group of individuals. The unemployed individual has the obligations to accept all programmes o¤ered and to be available for both non-subsidized and subsidized work. However, only a fairly low fraction of the unemployed participate in programmes during the passive period. Programmes are categorized into 4 types by the National Labour Market Authority: * Subsidized employment programmes with private employers. The individual is employed in the private sector for a 6 - 9 months period, and the employer is paid the subsidy corresponding to roughly 50% of the minimum wage. * Subsidized employment programmes with public employers. These programmes o¤er the individual temporary (6 - 12 months) jobs in public sector institutions. * Education/training programmes. These include all types (usually short-lasting) training programmes, based on the background of the unemployed individual concerned. * Other programmes, which include all programmes that cannot be classi…ed within one of the categories above. A variety of programmes is covered by this group, for example job search assistance, competence detection programmes, individual specialized job training (in case the unemployed individual cannot take ordinary training programmes), job rotation (when leave taken by employees is combined with a job training contract for unemployed workers), pool jobs (the jobs in the public sector of up to 3 years duration, having the goal to create more permanent jobs in the priority …elds), etc.

8 3. Econometric Model

The present paper is using the timing-of-events model for identifying treatment e¤ects in a duration model framework, developed by Abbring & van den Berg (2003). The timing-of-events model simultaneously models the transition rate out of unemployment, and the transition rate into the ALMPs. The model is intended to correct for non-random selection into programmes, with respect to observed as well as unobserved variables. Abbring & van den Berg (2003) show that with an assumption of 1) mixed proportional hazards and 2) a non-defective distribution of time until participation in ALMPs, given observed explanatory variables, the parameters of interest - say, the e¤ect of participation in ALMPs - are identi…ed non-parametrically. The implication is that there is no need for an exclusion restriction, that is, a variable, which appears in the selection equation, but which does not a¤ect the outcome variable, in this case the hazard rate out of unemployment. The intuition is that random variation in the timing of the event of participation in ALMPs separates the treatment e¤ect from the distribution of unobserved heterogeneity, which is assumed to be time-invariant.

Let Tu be a random variable, denoting the duration of an unemployment spell, and let Tp be another random variable, denoting the time from entry into unemployment until participation in the …rst ALMP. When we have Tp < Tu; the individual participates in an ALMP during the unemployment spell. If Tp Tu, then Tp is censored, and the individual didn’t participate in  an ALMP before Tu.

Let X(t) be a vector of observed exogenous explanatory variables, and let Vu, and Vp =

(Vp1;Vp2;Vp3;Vp4), denote the unobserved variables, possibly a¤ecting the exit rate out of un- employment and the entry rates into the four di¤erent types of ALMPs. The hazard into ALMPs is the sum of four cause-speci…c hazard rates, one for each type of ALMP: 4 p(tp x(tp); vp) = pi(tp x(tp); vpi): (1) j j i=1 X Each of these cause-speci…c hazards is assumed to be of the mixed proportional hazard type,

pi(tp x(tp); vpi) = pi(tp) exp x(tp) + vpi : (2) j pi 

9 Now we de…ne two time varying vectors of indicator variables, d1(t) and d2(t): d1(t) is a 4 1  vector, where the i th element takes the value 1 if an individual participates in an ALMP of type 0 i at time t and takes the value 0 otherwise. Note that at most one element of d1(t) can take the value 1 at time t. Similarly, the i th element of d2(t) (which is also 4 1) takes the value 1 0  if an individual has completed an ALMP of type i during the last 26 weeks (the implication is that we only allow ALMPs to a¤ect the hazard rate out of unemployment up to 26 weeks after completion). Assuming once again a mixed proportional hazard rate, the hazard rate out of unemployment in our model is speci…ed as

u(tu x(tu); d1(tu); d2(tu); vu) j

= u(tu) exp [x(tu) u + d1(tu)1 + d2(tu)2 + vu] : (3)

The parameter 1 here measures the locking-in e¤ect, while 2 - the post-programme e¤ect. In the estimations performed below, we will allow for separate e¤ects of programmes that start in di¤erent years, but for expositional convenience, this interaction - between participation and completion indicators on the one side and year dummies on the other - has been ignored.

The timing-of-events model takes into account potential endogeneity of d1(t) and d2(t) by allowing for correlation between the two unobserved components, Vu and Vp. That is, this method allows for selection of unobservables as well as observed explanatory variables.

We de…ne Cu as an indicator variable that takes the value 1 when the unemployment spells are completed and 0 for right censored unemployment spells, and so the contribution to the likelihood function of an individual with J unemployment spells, given observed and unobserved characteristics, is:

J 1 tpj

exp p(s x(s); vp)ds u(t x(t); d1(t); d2(t); vu)dt ; (4)  2 j j 3 Z0 Z0 4 5

10 and the likelihood function then can be expressed as:

= (vu; vp)dG(vu; vp); (5) L L ZZ where G( ; ) is a bivariate distribution function for (vu; vp).   The expected duration of an unemployment spell may be calculated as

1 E[Tu x; d ; d ; v ] = S(t x; d ; d ; v )dt; (6) j 1 2 u j 1 2 u Z0 where the time-variation in the explanatory variables has been ignored for analytical conve- nience.

3.1. Identi…cation

Identi…cation in the timing-if-events model is based on two assumptions, as mentioned above: a ’distributional’assumption, requiring the hazard rates to be speci…ed as mixed proportional hazards, and a ’no anticipation’ assumption, which implies that the individual is allowed to know the distribution of time until programme participation and the distribution of programme types, but not the exact moment, at which he will participate. Typically anticipation about the activation can occur after the meeting with a case worker, i.e. after a job planning meeting. ALMP participation then usually starts after a few weeks2. Thus the person knows about the activation only a few weeks in advance. If the time span between the earliest moment at which anticipation can occur and the moment of activation is relatively short and if the anticipatory e¤ect is not very large, estimation results may be fairly insensitive to the assumption of no anticipation.

2 Before the reform of Labour Market Policies in 1994, participation in ALMPs in Denmark led to renewed eligibility for UI bene…ts. Thus, the date of ALMP activation could be determined by the date of expiration of bene…ts entitlement, i.e. a no anticipation assumption could be violated. However, after the reform in 1994 only a full time employment of 26 weeks lets to renew bene…t eligibility.

11 3.2. Parameterization

We assume all baseline hazard rates to be piecewise constant (that is j(t) = exp( jm), m =

1; :::; Mj , where Mj is the number of intervals for baseline hazard j). The following cut-o¤ points for the intervals are used for all hazard rates (the unemployment duration and time until programme participation are both measured in weeks): 4, 13, 26, 39, 52, 65, 78, 91, 104 and 156. With such a parameterization, it is straightforward to show that the expected duration of an unemployment spell is equal to

M u 1 E[Tu x; d ; d ; v ] = j 1 2 u hm(x; d ; d ; v ) m=1 i 1 2 u X P ( m 1 < Tu  m x; d1; d2; vu) ; (7)   j

m where the  m denote the cut-o¤ for the intervals, and hi is the value of the hazard rate in interval m, and P ( m 1 < Tu  m x; d1; d2; vu) is the probability that an individual leaves  j unemployment in the m’thinterval. For the mixture distribution, we apply a discrete distribution with two points of support 1 2 1 2 for each of the marginal distributions of the unobserved variables. Let (vu; vu) and (vpi; vpi), i = 1; 2; 3; 4, be the mass-points of Vu and Vpi, respectively. The associated probabilities are then:

1 1 1 1 1 P1 = P r(Vu = vu;Vp1 = vp1;Vp2 = vp2;Vp3 = vp3;Vp4 = vp4); 2 1 1 1 1 P2 = P r(Vu = vu;Vp1 = vp1;Vp2 = vp2;Vp3 = vp3;Vp4 = vp4); 1 2 2 2 2 P3 = P r(Vu = vu;Vp1 = vp1;Vp2 = vp2;Vp3 = vp3;Vp4 = vp4); 2 2 2 2 2 P4 = P r(Vu = vu;Vp1 = vp1;Vp2 = vp2;Vp3 = vp3;Vp4 = vp4);

4 with 0 Pi 1 for i = 1; ::::; 4, and Pi = 1. Note that the unobserved heterogeneity   i=1 terms are restricted to be perfectly correlatedX in the four cause-speci…c hazard rates into pro- grammes. This is also called a factor-loading speci…cation. It restricts the correlation between

Vpi and Vpj to be either 1 or 1 if i = j. Such a parameterization of the unobserved hetero- 6 geneity distribution is rather restrictive; however, this ’factor loading speci…cation’ has been chosen for computational reasons, i.e. to restrict the number of unknown parameters and to limit the computational burden of the estimation of the model.

12 The correlation between Vu and Vp is unrestricted, which is important, since this is the correlation, which is intended to correct for selection of unobservables. We normalize the 1 distribution of the unobservables by setting vj = 0 for all hazard rates. This is done instead of normalising e.g. the mean of the mixture distribution to one.

4. Data

There have been made 28 data sets - 14 for men and 14 for women in di¤erent Danish counties (excl. Bornholm) - for the research. The data was extracted from an event history data set developed by the Danish National Labour Market Authority. The event histories are based on the administrative registers, which record and govern the payments of public income transfers, as well as the register, in which the employment agencies record the unemployed’sparticipation in ALMPs. Using these event histories, constructed by the NLMA itself, the employment agencies determine the risk that an individual becomes long-term unemployed (Hammer et al., 2004), so in this respect not only the underlying information, but also the event histories themselves are considered to be a very reliable data source. The data used in this paper covers the period from January 1, 1999 to December 31, 2004. The records are done on a weekly basis and include all time periods when the unemployed has received a public income transfer. Since this data is used for administrative purposes, it is frequently updated and therefore not merged with other registers, containing information on such variables as education and work experience. In this paper we concentrate on the unemployment spells of workers who are eligible for UI bene…ts, since the information available for UI recipients is of a much higher quality than for social assistance recipients. An unemployment spell means the period, in which the individual is either openly unemployed or participates in an ALMP. If there can be found four consecutive weeks out of open unemployment, when a person does not receive any other public income transfer, then he is treated as having found a job. If an individual has more than four weeks out of unemployment receiving other transfers, the unemployment spell is characterised as right censored. Periods out of unemployment, shorter than four weeks, which usually include paternity leave

13 (two weeks), holiday periods (typically up to three weeks), or short periods of sickness bene…t payments, are excluded from the unemployment spell. The samples consist of 25% of observations in each data set, drawn randomly3. The age of the individuals in the sample is limited between 25 and 59 (both included). Another - much stricter - unemployment policy is applicable to the individuals, younger than 25 years, while for those over 59 special rules (in this case - much milder) are valid as well, so both of these age groups are excluded from our research. The temporary lay o¤ unemployment is eliminated by excluding from our samples all un- employment spells, lasting less than four weeks (note: about 40% of the unemployment spells belong to temporary unemployment, more than 90% of them lasting less than four weeks), since the ALMPs are not used in the case of a short term unemployment4.

Only 36.8 % of the individuals in the data set experience a single UI bene…t spell, while the rest persons are subject to repeated unemployment (see Table 2). The average time span between the …rst and the second unemployment spells is 37.3 weeks, and the time span declines for subsequent spells. This is in part due to the …nite observation window.

Table 2. Frequencies of UI bene…t spells

Individuals with a single UI spell, % 36.8 Individuals with two UI spells, % 21.6 Individuals with three UI spells, % 14.0 Individuals with four or more spells, % 27.7 Average time between 1st and 2nd spell, weeks 37.3 Average time between 2nd and 3rd spell, weeks 30.4 Average time between 3rd and 4th spell, weeks 27.2

This paper, however, assumes that programme participation in past unemployment spells has no e¤ect on current unemployment duration, i.e. the post-program e¤ects do not carry

3 Using more observations would result in less statistical uncertainty in the estimations. However, this was not feasible for computational reasons. 4 Excluding from the data the spells lasting less than 4 weeks induces a left truncation problem and thus implies a correlation between unobserved and observed covariates. However, most of these spells are recall unemployment, which is a completely di¤erent stochastic process and which is not a target of Danish Active Labour Market Policies. Including these short-lasting spells into the data could lead the e¤ects of ALMPs to be underestimated. In any case, just conditioning on the event of unemployment leads to a selection problem.

14 over between the subsequent spells, which is a fairly strong assumption5. This issue has got some attention in the existing literature, for example, Røed & Westlie (2007) model the intrinsic duration dependence as a function of three factors: i) the overall unemployment exposure during the four year period prior to the current spell, ii) the time that elapsed from the end of the last completed spell to the start of the current spell, and iii) the duration of the ongoing spell. Description of the sample is given in Table A.1 in the Appendix.

4.1. Explanatory Variables

A number of explanatory variables is used for the research. The three age groups - AGE30-39, AGE40-49 and AGE50-59 - represent the age of the unemployed. There is a set of indicators, showing the UI fund membership: UI FUND CONSTRUCTION, UI FUND MANUFACTUR- ING, UI FUND TECHNICIANS, UI FUND TRADE, UI FUND CLERICAL, UI FUND ACA- DEMICS, OTHER UI FUND, and UI FUND SELF-EMPLOYED. Some of the UI funds exist based on the industry, while others are based on the educational achievements of the mem- bers. For example, UI FUND MANUFACTURING mainly insures unskilled workers of the manufacturing industry, and UI FUND ACADEMICS covers UI funds, which insure academ- ically educated workers. Thus, the UI fund membership to some extent represent educational attainment, and to some extent - past occupation. An indicator of the marital status and two indicators, representing the country of origin, have been included (the reference category is native Danes). The indicator that an individual lives alone is SINGLE. IMMIGRANT FROM DC covers the …rst and second generation immigrants from developed countries, while IMMIGRANT FROM LDC - immigrants from less developed countries. There is an access to the past labour market history information. We have the data about the fraction of time spent on public income transfers for each of the past …ve years; we know as well how many spells, receiving public income transfers, a person has experienced within the past …ve years. But only the information from the two last years, TRANSFER DEGREE LAST YEAR, TRANSFER DEGREE TWO YEARS AGO, # TRANSFER SPELLS LAST

5 However, we have to some extent corrected for past unemployment, i.e. we have included the variables about the fraction of time spent on public income transfers and about the number of spells of public income transfers for the past two years.

15 YEAR, and # TRANSFER SPELLS LAST 2 YEARS was signi…cantly di¤erent from zero in preliminary explorations, and therefore only information from these past two years before the unemployment spell is used. Finally, we also have access to information about sickness spells. The data from the last two years has been used in our research, for the same reason as given above.

5. Meta Analysis Framework

To summarize the e¤ects of ALMPs in Danish counties the framework of meta-analysis – a technique for analyzing and summarizing the results of di¤erent studies, each of which is focused on the same question –is used. The meta analysis techniques have been developed and applied widely in the area of medical and natural sciences. Application of these techniques in Economics has been developed by Phillips (1994), Card and Kruger (1995), Ashenfelter and others (1999), van der Sluis and others (2004), van der Sluis and others (2005), Kluve and Schmidt (2002) and Kluve (2006). In van der Sluis and others (2005) the meta-analysis of the impact of education in developing economies has been carried out, using ordinary least squares and ordered probit. The meta-analysis here is de…ned as ’aquantitative tool that is applied to synthetize previous research …ndings that share common aspects that can be addressed statistically’. The idea to use meta-analysis to measure the e¤ectiveness of European ALMPs was …rst implemented by Kluve and Schmidt (2002), who summarized a total of 53 European active labour market programmes. It was further developed by Kluve (2006), when he summarized the e¤ects of European ALMPs from all European evaluation studies, available to that date (95 di¤erent evaluation studies). The basic idea of meta-analysis in these studies was to construct and analyse a data set in which each observation represented a particular programme evaluation. The motivation for such an analysis was its ability to identify systematic di¤erences across di¤erent types of ALMPs, while controlling for other factors, like economic conditions, during the period of the evaluation. Various programmes and evaluation methods are used in order to estimate the e¤ects of European ALMPs. Thus, talking about the problems of implementing meta-analyses, the dif- ferences among the evaluation studies tend to be an important issue. Firstly, there is a variation

16 among the ALMPs in di¤erent countries - the programmes di¤er in their design and in their focus on di¤erent target groups. Secondly, the institutional and economic environment di¤ers among the countries and thirdly, there are disparities in the evaluation design and estimation techniques. In our research these problems are not present, since the e¤ectiveness of ALMPs in Danish counties is measured using the same methodology - the timing-of-events model for identifying treatment e¤ects in a duration model framework, described in Section 3. The types of ALMPs do not di¤er among the counties either.

We have two goals of meta-analysis. Firstly, we need to answer, which of the Danish ALMPs have shown the best performance in the and in the whole country. In order to be able to get the answers, we need to summarize the annual results of the timing-of events models, estimated separately for men and women in each of the 14 counties in Denmark. We calculate the weighted averages of the e¤ects of the programmes for men and for women in the country and at county levels. We have 28 samples (for men and for women in 14 counties). However, weighting the e¤ects of the programmes by sample sizes, as it is common in statistical analyses, is not an optimal approach in our case, since the estimated e¤ects are based on those who participate in programmes in di¤erent counties in di¤erent years. We therefore take the numbers of individuals, activated in each of the ALMPs in a given calendar year as the optimal weights for our analysis. Thus, we have:

ESprogramme;i;t Ni;t  i;t ESprogramme = X ; Ni;t i;t X

where ESprogramme is the average e¤ect of a programme (locking-in e¤ect, post-programme e¤ect or net-e¤ect);

ESprogramme;i;t is the e¤ect of a programme in county i in calendar year t;

Ni;t is the number of unemployed individuals in county i, activated in the programme in calendar year t;

17 Ni;t is the total number of the individuals, activated in the programme. i;t X To …nd the country-level e¤ect of the programme we calculate a weighted average of 84 observations (from 14 counties and 6 calendar years of research). In the same way we …nd the average county-level e¤ects as weighted averages of 6 annual observations.

The second goal of our meta-analysis is to obtain a quantitative assessment of the factors (i.e. economic conditions, composition of the work force and the labour market, intensity of programme use), which are perceived to be associated with the e¤ect of ALMPs. We construct two data sets - one for men and another for women - each of them having 84 observations (14 counties * 6 calendar years). The dependent variable of interest is given by the net e¤ect of the programme evaluated6. The net e¤ect is de…ned as the impact on the expected duration of unemployment for a standard person, and here we use the same standard person in each region. Participation is assumed to take place from week 52 in the unemployment spell. The net e¤ect of each programme is related to a set of county-level variables in a linear regression model. "Activation, %" is de…ned as the fraction of the unemployed individuals, activated in the particular type of ALMPs (individuals activated in all 4 ALMPs types = 100%). ‘Low-skilled’is the fraction of persons, employed as ’lønmodtagere grundniveau’in a county in a given year. The fraction of the low-skilled workers is steadily decreasing. It went down from 41.8% in 1999 to 34.9% in 2004 for men and from 44.8% to 40.4% for women. The biggest reduction in the fraction is noticeable in 2004. The county level …gures varied as well: from the average of 32.5% (in Copenhagen and Frederiksberg region) to 43.7% (in Viborg county) for men and from 38.4% (in Copenhagen and Frederiksberg) to 46.4% (in Storstrøms county) for women. ’Manufacturing, %’, ’Private sector, %’and ’Self-employed, %’show the fractions of indi- viduals, employed in manufacturing, in private sector, and the self-employed individuals. In the manufacturing sector work about 21% of men, while for women this …gure comes up to about 11.5%. The lowest fraction of manufacturing tends to be in the area of capital and around the capital (Copenhagen and Frederiksberg municipalities, Copenhagen and Frederiksborg coun-

6 The fact that meta-regression analysis uses dependent variable, which is a result of the …rst step estimations, makes the reported standard errors to be too low. However, the net e¤ects in the …rst step estimation are statistically signi…cant, thus the standard errors in the meta-regression are reliable.

18 ties), while the biggest - in the Central Jutland (Ringkøbing, Viborg and Vejle counties). 75.3% of men work in the private sector, while for women this …gure comes up to only 45.5%. The highest percentages of the employed women work in social institutions (21.2%), health and education sectors (9.4% in each). Among men the most popular are the metal industry (9.6%) and construction (10.4%). These …gures di¤er among the counties from 63.7% employed men in the private sector in Copenhagen and Frederiksberg region to 82.5% in Ringkøbing county; and from 37.3% (Storstrøm) to 51.6% (Copenhagen county) for women. 10.5% of the working force of men and 3.9% of women are self-employed. The biggest share of self-employed men is in Viborg (14.4%); of self-employed women - Storstrøm county (4.7%). ’Unemployment, %’is regional unemployment rate, ’>50 years old, %’- the fraction of those older that 50 years among the 30-59 years aged unemployed individuals, that equals to about 32 - 33% for both men and women. ’Duration, weeks’is the expected unemployment duration (without activation), measured in weeks. In the standard notation of programme evaluation, this variable is E[T0], the outcome in the absence of programme participation, where E[T1] is the outcome in the presence of programme participation. The coe¢ cient to this variable will be informative about the trade-o¤ between equity and e¢ ciency in active labour market policy; if those with large unemployment duration are also those, whose unemployment duration is reduced the most, then there is no trade-o¤, while in the opposite case, there is a controversy between the goals of equity and e¢ ciency. Unemployment duration is 1.5 - 2 times longer for women than for men and di¤ers from 8.5 weeks (men) and 17.6 weeks (women) in Viborg to 22.5 weeks (men) and 34.6 weeks (women) in Copenhagen and Frederiksberg municipalities. County level statistical information about unemployment is available in Tables A.8.1. - A.8.2. in the Appendix. Appendix Tables A.9.1. - A.9.8. present the fractions of low-skilled in the working population of men and women, the percentages of men and women occupied in manufacturing, the fractions of men and women in the private sector as well as the fractions of those self-employed, Tables A.10.1. - A.10.2. - the age structure (by %) of the 30-59 years old unemployed individuals. The analysis is conducted using a linear regression framework. To control for any permanent features of di¤erent years, that may in‡uence the net e¤ect of the ALMPs, we include year

19 dummies. We use 1999 as the omitted year in the base category, i.e. the annual e¤ects are judged relative to 19997. Section 6 presents and discussed the estimation results, while section 7 - the results of the meta analysis.

6. Estimation Results

This section covers the results of the analyses and attempts to answer, which of the ALMPs has shown the best performance for the employability of the unemployed men and women and for their unemployment duration in di¤erent counties of Denmark, and in the whole country. The timing-of events model is estimated separately for men and women, and for each of the 14 counties in Denmark (because of few observations, we have not estimated the model for the county of Bornholm). In addition, the models are speci…ed with an interaction between year and programme participation and completion indicators, such that the e¤ects of programmes are estimated separately by calendar year. The results are presented in the Tables 4-8. Table 4 shows the average programme e¤ects in the country, while county-level e¤ects are presented in Tables 5-8. The best results among the four types of the ALMPs, achieved by the private job training, are marked in bold in Table 4, while in Tables 5-8 we marked in bold the three most favourable (or least harmful) results, produced by the programmes in Danish counties. The e¤ects have been averaged using the methodology explained in Section 5. ’Locking-in e¤ect’and ’Post-programme e¤ect’in the tables show the coe¢ cients estimated, that is, the exponential function of the reported parameter gives the multiplicative impact on the hazard rates. The ’Nete¤ect’shows the e¤ect of the programmes on expected unemployment duration. The net e¤ect is calculated for a ’standardperson’with close to average characteristics for each of the samples used. Participation is assumed to take place after 52 weeks of unemployment, and last for 26 weeks (private and public job training), 16 weeks (education) and 8 weeks (other ALMPs).

7 One of the goals of the e¤ect-time-interaction is to re‡ect that the e¤ects of programmes may depend on labour demand conditions. Secondly, the composition of ALMPs may di¤er between the years. An option to interact the e¤ects of programmes with the current time instead is not possible, since the estimations are done in two steps, and a linear regression framework doesn’tallow for time varying indicators.

20 Calculating the net e¤ect by comparing expected durations with a treatment occurring after one year with expected durations given that no treatment occurs implies that this summary measure depends on two factors, i.e. on the probability that unemployment spells last at least one year and on the e¤ect of treatment given that the spell is long enough to ensure that a treatment actually occurs. This choice, however, is based on the average length of the time span since the start of unemployment until ALMP activation of the individuals in the data set, which is found to be about 52 weeks. Looking into ALMP activations in the di¤erent intervals of the UI bene…t spell (see Table 3) reveals that only 5.7% of unemployed persons were subject to ALMP activation during the …rst half year of unemployment, while after one year spent on UI bene…t the probability of programme participation increases to 34.4%.(and remains close to that level thereafter).

Table 3. ALMP Activations in Di¤erent Intervals of UI bene…t spell8

Months of UI Bene…t Spell 1-6 7-12 13-18 19-24 >24 Private Job Training 0.3 1.7 2.1 2.4 2.0 Public Job Training 0.5 3.2 5.7 7.2 7.5 Education 3.2 12.9 18.9 19.0 16.2 Other ALMPs 1.6 5.3 7.5 6.8 6.9 Total 5.7 23.1 34.3 35.5 32.7

Going into more details we should note that the expected duration of an unemployment spell can be divided into time intervals before and after ALMP activation, i.e.

tp 1 E(T ) = S(t)dt + S(t)dt;

Z0 Ztp

and the relative sizes of ALMPs e¤ects in our analysis are expected to be the same (since the component of the integral until activation is identical for programme participants and none- participants).

8 Note that here we present the ratio between the number of persons activated in ALMPs in each interval of the UI bene…t spell and the number of unemployed persons at the start of each interval. It does not sum up to 100% and should not be confused with the ratio between the number of persons activated in ALMPs in each interval of the UI bene…t spell and the number of persons activated in ALMPs during the period of observation.

21 Selection of the length of ALMPs is based on a similar motivation, i.e. on average lengths of the treatments of the unemployed individuals in the data set.

The year and county speci…c e¤ects of the programmes are shown in the Appendix Tables A.2.1 - A.6.8. Tables A.2.1. - A.2.16. here present the locking-in e¤ects, Tables A.3.1. - A.3.16. - the post-programme e¤ects. The annual county-level results of unemployment durations (with and without programme participation) and the net-e¤ects of the ALMPs are shown in Tables A.4.1. - A.4.9 and Tables A.5.1. - A.5.89, while Tables A.6.1. - A.6.8. cover the percentage changes in unemployment duration, resulting from participation in the programmes.

Table 4. Average Locking-in, Post-programme and Nett E¤ects Of ALMPs in Denmark, 1999 - 2004

Locking-in e¤ect, % Post-prog. e¤ect, % Net e¤ect (weeks) Private job training Men -17.5 60.0 -0.21 Women -12.2 57.6 -0.85 Public job training Men -58.7 -8.4 1.22 Women -55.6 0.5 1.99 Education Men -59.8 -6.6 2.01 Women -68.4 -5.1 1.93 Other ALMPs Men -25.8 -17.0 0.45 Women -40.8 -11.2 1.00

From Table 4, where the columns for the locking-in and post programme e¤ects show the percentage change in the hazard rate, we observe that only one group of ALMPs - private sector employment programmes - reduced unemployment duration, by 0.85 weeks for women and by 0.21 weeks for men. The small locking-in e¤ect of the programme was followed by a positive post programme e¤ect, when the hazard rate out of unemployment was increased by

9 Standard errors in tables A.4.1. - A.4.9 and A.5.1. - A.5.8 are calculated by drawing 500 parameter vectors from a multivariate normal distribution with the estimated parameter mean and variance-covariance matrix. Based on these estimates the expected duration and net e¤ect measures are calculated 500 times. The standard errors come about as the empirical standard deviation of the 500 estimated e¤ects.

22 60% for men and by 57.6% for women. For education and other ALMPs we …nd both locking-in and post-programme e¤ects to be negative, which lead to the positive net programme e¤ects (the programmes increased unemployment duration). The post-programme e¤ect of public job training for women found to be slightly positive (0.5%), however this does not compensate the dramatic locking-in e¤ect. The worst performance was shown by education, which increased unemployment duration of both men and women by about 2 weeks.

The county-level e¤ects of private sector employment subsidy programmes are presented in Table 5. The table shows that the estimated average post-programme e¤ect is positive for both men and women in all counties, and in most counties the post programme e¤ect exceeds the locking-in e¤ect, and expected unemployment duration is reduced (except Viborg county for men, and Frederiksborg, Western and Ribe counties for women). For both men and women, we …nd the best performance of private sector employment programmes in Roskilde and Northern Jutland. It is worth mentioning that the programme e¤ects also look very favourable in Copenhagen and Frederiksberg municipalities for women (1.87 weeks decrease in unemployment duration) and in Frederiksborg county for men (0.40 weeks decrease). Overall, for private sector employment programmes, we would thus conclude that there is a tendency that these programmes perform better in the area around the capital, the counties with the lowest unemployment rate, and in Northern Jutland, where the unemployment rate is one of the highest in the country. The estimated e¤ects of the other 3 groups of ALMPs (see Tables 6-8) are not nearly as positive, in fact only few programmes have sizeable positive post-programme e¤ects and they nearly all have large locking-in e¤ects, and increase unemployment durations in all except one case, especially public sector employment programmes and education programmes. The application of the public sector employment programmes to men (Table 6) results in negative post-programme e¤ects in most of the counties (with the exception of very slightly positive e¤ects in Copenhagen and Frederiksberg municipalities, and ). For women, however, the estimates of post-programme e¤ects are slightly or moderately positive in half of the counties, with the best result - 19.3% - found in Copenhagen and Frederiksberg. In spite of the positive post-programme e¤ect, Copenhagen and Frederiksberg region experienced

23 the most harmful consequences in the increasing of unemployment duration - by 2 weeks for men and 2.28 weeks for women. The unemployment duration increased signi…cantly in Copenhagen and Roskilde counties. The best results are found to be in Aarhus, where the employment probabilities for women and men increased by 15% and 3.5% respectively at the end of programme participation. But even these moderate positive e¤ects were not nearly large enough to compensate for the locking- in e¤ects and to reduce unemployment duration. The education programmes (Table 7) on average gave the worst impacts for both men and women. In light of the fact that this is the most frequently used type of programmes, this is of course very disappointing. With the exception of Funen, where the post-programme hazard rate out of unemployment for men increased by 7%, within the whole country the post-programme education e¤ect was negative for women and negative or only very slightly positive (not larger than 4%) for men. However, the probability of leaving unemployment during the education period was reduced by 46-70% for men while for women it decreased by even more - 64-72%. The biggest increases in unemployment duration - 2.7 and 2.4 weeks (both men and women) - have been found in Copenhagen and Frederiksberg as well as in Copenhagen county, the lowest - 1.3 weeks - in Viborg. ‘Other active labour market programmes’(Table 8) are found to be ine¤ective. The locking- in e¤ects here are milder than in the cases of public job training and education programmes. But on the other hand these (other) programmes do not produce positive post-programme e¤ects either. Moreover, there are only two counties - Frederiksborg and Roskilde - that are experiencing a slightly positive post-programme e¤ect to the hazard out of unemployment of women, while for men in all the counties these e¤ects are negative. Only for women in Roskilde county a slight - 0.16 weeks - decrease in unemployment duration is observed, while in the other counties duration is increased for both men and women (with the highest increase in Copenhagen and Frederiksberg region, and Copenhagen county as well as in Nothern Jutland for women, and in Roskilde for men).

24 Table 5. E¤ects of Private Job Training Programmes in Danish Counties (excl. Bornholm) in 1999-2004 Locking-in e¤ect, % Post-prog. e¤ect, % Net e¤ect (weeks) Men Women Men Women Men Women Copenhagen and Frederiksberg -25.87 -17.83 50.31 91.13 -0.05 -1.87 Copenhagen county -18.93 -9.03 40.31 42.50 -0.19 -0.92 Frederiksborg county -1.13 -36.50 46.06 37.00 -0.40 0.32 Roskilde county 6.92 12.07 54.06 56.13 -0.81 -1.51 Western Zelland county -0.90 -34.70 26.84 7.81 -0.19 1.05 Storstrøms county -23.33 -31.16 64.56 56.50 -0.13 -0.35 Funen county -19.28 -15.21 61.18 79.39 -0.11 -0.64 Southern Jutland county -0.44 -15.02 28.78 58.40 -0.35 -0.74 -35.78 -33.62 68.64 10.65 -0.02 0.86 Vejle county 3.10 0.92 68.42 64.26 -0.20 -1.12 Ringkøbing county -11.81 18.35 55.28 38.06 -0.13 -1.19 Århus county -18.91 -5.46 82.09 70.60 -0.30 -1.16 Viborg county -40.03 -18.10 44.76 40.41 0.12 -0.15 Northern Jutland county -24.05 -2.19 93.62 83.08 -0.34 -1.67

Table 6. E¤ects of Public Job Training Programmes in Danish Counties (excl. Bornholm) in 1999-2004 Locking-in e¤ect, % Post-prog. e¤ect, % Net e¤ect (weeks) Men Women Men Women Men Women Copenhagen and Frederiksberg -61.11 -57.46 1.77 19.34 2.00 2.28 Copenhagen county -57.96 -54.84 -12.49 -11.21 1.86 2.42 Frederiksborg county -61.28 -54.19 -12.21 0.36 1.29 1.89 Roskilde county -71.79 -50.20 -15.60 -16.41 1.85 2.26 Western Zelland county -66.71 -60.77 -2.52 -2.86 1.37 2.46 Storstrøms county -58.28 -55.37 -29.41 -2.19 1.26 1.89 Funen county -70.82 -76.52 -17.79 2.27 0.84 2.24 Southern Jutland county -54.16 -59.11 -5.96 -28.50 0.90 2.48 Ribe county -50.67 -60.96 -13.53 -5.80 0.55 2.40 Vejle county -53.62 -53.06 -0.42 -3.01 1.04 1.75 Ringkøbing county -62.35 -54.59 -30.74 11.52 0.67 1.65 Århus county -48.59 -28.15 3.45 14.92 0.77 1.37 Viborg county -62.82 -57.36 -11.01 6.43 0.58 1.37 Northern Jutland county -47.57 -52.01 -1.16 3.02 0.89 1.90

25 Table 7. E¤ects of Education Programmes in Danish Counties (excl. Bornholm) in 1999-2004

Locking-in e¤ect, % Post-prog. e¤ect, % Net e¤ect (weeks) Men Women Men Women Men Women Copenhagen and Frederiksberg -58.11 -66.95 -6.61 -5.11 2.70 2.70 Copenhagen county -53.08 -66.08 -6.88 -8.28 2.43 2.43 Frederiksborg county -52.80 -68.91 -11.83 -8.07 1.82 1.83 Roskilde county -47.22 -65.04 3.89 -3.58 2.01 1.99 Western Zelland county -45.67 -64.04 -6.44 -8.98 1.92 1.92 Storstrøms county -61.77 -69.52 3.68 -8.32 1.54 1.54 Funen county -57.77 -65.02 7.00 -1.49 1.49 1.49 Southern Jutland county -58.70 -72.45 -7.85 -7.09 1.85 1.87 Ribe county -56.52 -72.17 -0.27 -3.68 2.01 2.03 Vejle county -67.75 -67.71 -2.02 -7.52 1.93 1.93 Ringkøbing county -70.10 -70.72 -14.31 -14.12 1.83 1.86 Århus county -65.41 -68.90 -11.77 -6.55 1.78 1.79 Viborg county -65.85 -69.55 -4.56 -3.12 1.31 1.32 Northern Jutland county -64.77 -71.32 -45.34 -3.32 1.73 1.74

Table 8. E¤ects of Other ALMPs in Danish Counties (excl. Bornholm) in 1999-2004

Locking-in e¤ect, % Post-prog. e¤ect, % Net e¤ect (weeks) Men Women Men Women Men Women Copenhagen and Frederiksberg -18.98 -44.72 -16.55 -18.68 0.60 1.70 Copenhagen county -27.49 -38.27 -17.65 -9.85 0.61 1.37 Frederiksborg county -53.54 -48.48 -13.15 5.72 0.54 0.41 Roskilde county -35.59 -9.48 -31.82 13.75 0.87 -0.16 Western Zelland county -23.20 -25.00 -10.95 -6.19 0.30 0.43 Storstrøms county -23.60 -31.05 -21.46 -20.16 0.39 0.87 Funen county -38.84 -41.84 -17.20 -10.61 0.25 0.60 Southern Jutland county -9.10 -10.65 -17.30 -7.50 0.24 0.18 Ribe county -28.85 -63.18 -13.44 -2.01 0.15 0.83 Vejle county -34.24 -34.49 -9.62 -8.21 0.30 0.58 Ringkøbing county -26.21 -53.86 -22.39 -10.94 0.19 0.98 Århus county -7.58 -32.54 -17.06 -5.93 0.23 0.55 Viborg county -32.81 -50.31 -21.61 -14.50 0.22 0.73 Northern Jutland county -37.03 -57.03 -15.27 -16.23 0.35 1.25

26 Based on the results above we conclude that only in one of the four groups of Danish active labour market programmes - private job training –is the post-programme e¤ect large enough to counteract the locking-in e¤ect in many of the counties, and thus to reduce expected unem- ployment duration, while the participants in the other three types of programmes experience increasing unemployment duration. The public job training programmes produced in some cases a mild or moderate positive e¤ect in the post programme period, but a large locking-in e¤ect during the programme period, and so unemployment duration increased. The ‘Other ALMPs’group also showed negative e¤ects (except for women in Roskilde), while the poorest performance by far was achieved by education programmes - with dramatic locking-in e¤ects, negative or slightly positive post-programme e¤ects, and increases in unemployment duration. There is a tendency that private sector employment programmes perform better in the area around the capital. It is however important to stress also that these programmes were very favourable in Northern Jutland, where the unemployment rate is one of the highest in the country. The e¤ects of the other 3 groups of ALMPs, however, show di¤erent regional patterns. In terms of geography, public sector training, education and other ALMP programmes worked badly in the area of the capital and around the capital. The least-harmful results of these programmes experienced mid-Jutland - Aarhus and Viborg - counties, however even there unemployment duration increased.

7. E¤ects of Other Characteristics

This section brie‡y presents duration dependence and of the e¤ects of personal characteristics and unobserved heterogeneity to the hazard rates out of unemployment and to the di¤erent types of ALMPs. We use the Meta analysis technique presented in Section 5 to calculate the weighted averages of the e¤ect of the covariates for men and women in the country10, taking the sample sizes (see Appendix Table A.1.) as the optimal weights. The results for men are presented in Table 9 while Table 10 covers the e¤ects of the other covariates for women. Results signi…cant at 1% (5%) level are marked in bold (italic).

10 Since the e¤ects of ALMPs is the primary interest of the paper, this section presents only the country level e¤ects of other characteristics due to expositional reasons. The county level results, however, are available to the reader upon request.

27 Table 9. E¤ects of Other Characteristics on Hazard Rates of Men in Denmark, 1999-2004

From UI To PrJob To PubJob To Educ To Other Baseline Hazard, % 0-4 weeks 28.47 4-13 weeks 11.27 0.13 0.06 0.76 0.38 13-26 weeks 8.64 0.20 0.10 1.03 0.52 29-39 weeks 6.74 0.68 0.45 2.29 1.19 39-46 weeks 0.54 0.49 2.88 1.49 46-52 weeks 5.84 0.83 1.11 4.78 2.07 52-58 weeks 1.11 1.58 7.13 3.11 58-65 weeks 4.86 1.09 2.00 5.29 3.22 65-78 weeks 4.47 0.65 1.77 6.36 2.46 78-91 weeks 4.16 0.88 1.76 5.07 2.46 91-104 weeks 4.09 0.97 1.55 3.55 2.48 104-156 weeks 3.64 - 0.99 1.96 2.37 >156 weeks 3.93 -- 1.81 - Personal Characteristics, % 30-39 years old -6.4 -32.8 -46.6 -21.6 -19.3 40-49 years old -16.4 -48.4 -26.7 -20.5 -16.6 50-59 years old -38.4 -65.9 -24.4 -37.7 -29.3 Single -20.5 -16.1 23.6 3.7 17.2 Immigrant from DC -15.2 18.9 -14.0 49.7 -12.4 Immigrant from LDC -43.2 -10.6 -20.9 71.9 14.0 UI fund Construction 12.4 -27.5 -52.7 -39.5 -10.8 UI fund Manufacturing -3.5 -37.0 -37.7 -19.0 0.3 UI fund Technicians -23.7 -15.1 -53.8 1.2 -11.9 UI fund Trade -31.9 -42.6 -61.3 -9.8 2.9 UI fund Clerical -8.8 -74.4 -65.1 -18.9 -15.5 UI fund Academics -15.3 -41.7 -53.8 -4.8 7.6 Other UI fund -21.3 -38.3 -64.2 -17.2 -9.6 UI fund Self-employed -51.5 -55.8 -78.7 -8.0 -20.6 Transfer degree last year -3.5 -1.5 32.8 19.2 10.1 Transfer degree 2 years ago -12.1 -24.7 24.9 28.5 -8.2 Sickness last year -32.9 -36.5 -15.1 -1.5 -18.1 Sickness last 2 years 27.8 9.7 -24.5 -24.0 8.6 Transfer spells last year -1.3 -5.7 6.8 4.1 4.9 Transfer spells last 2 years -4.7 -20.0 2.8 8.7 -13.9 2 2 Unobserved Heterogeneity vu(%) vpi(%) -46.7 -49.3 79.4 38.0 29.3 1 1 1 2 P1 = P r(Vu = vu;Vpj = vpi) 0.36 P3 = P r(Vu = vu;Vp1 = vpi) 0.09 2 1 2 2 P2 = P r(Vu = vu;Vp1 = vpi) 0.33 P4 = P r(Vu = vu;Vp1 = vpi) 0.22

28 Table 10. E¤ects of Other Characteristics on Hazard Rate of Women in Denmark, 1999-2004

From UI To PrJob To PubJob To Educ To Other Baseline Hazard, % 0-4 weeks 24.66 4-13 weeks 5.32 0.09 0.08 0.53 0.35 13-26 weeks 3.82 0.13 0.11 0.61 0.41 29-39 weeks 3.13 0.32 0.36 1.88 0.71 39-46 weeks 0.40 0.47 2.19 1.26 46-52 weeks 2.96 0.76 0.65 3.42 1.69 52-58 weeks 0.58 0.93 4.72 2.35 58-65 weeks 3.08 0.67 0.86 4.78 2.67 65-78 weeks 2.66 0.57 1.03 4.17 1.93 78-91 weeks 2.52 0.52 0.79 3.33 1.81 91-104 weeks 2.51 0.60 0.87 2.99 1.59 104-156 weeks 2.48 0.50 0.50 2.49 1.66 >156 weeks 2.66 - 0.46 2.51 1.78 Personal Characteristics, % 30-39 years old 6.7 -36.7 -30.5 -4.0 -8.4 40-49 years old -6.7 -49.9 -2.4 17.0 8.8 50-59 years old -38.0 -73.2 1.9 -10.9 5.7 Single -12.5 9.7 26.3 18.6 30.7 Immigrant from DC 0.2 13.4 5.2 32.0 32.1 Immigrant from LDC -47.4 -21.7 -19.0 77.6 11.1 UI fund Construction 5.7 -21.8 -27.4 -19.2 6.5 UI fund Manufacturing -9.8 -49.6 -37.8 -6.8 -4.4 UI fund Technicians -3.7 -34.4 -64.3 -13.8 -36.9 UI fund Trade -3.9 -50.2 -47.3 -24.8 -18.5 UI fund Clerical 29.5 -87.0 -57.1 -33.6 -34.0 UI fund Academics 5.3 -47.7 -36.3 -22.6 -10.8 Other UI fund -1.5 -68.4 -36.4 -24.3 -29.5 UI fund Self-employed -50.3 -66.7 -62.2 1.9 -28.1 Transfer degree last year -5.7 12.2 8.8 11.1 -5.8 Transfer degree 2 years ago -6.6 -25.1 38.1 31.3 -3.3 Sickness last year -27.0 28.8 18.9 9.5 -12.9 Sickness last 2 years 36.4 -17.4 -37.7 -35.4 -9.1 Transfer spells last year -2.6 2.8 0.3 2.1 -6.7 Transfer spells last 2 years -3.3 -20.9 3.1 5.7 -7.7 2 2 Unobserved Heterogeneity vu(%) vpi(%) -53.5 -58.3 63.6 76.2 72.3 1 1 1 2 P1 = P r(Vu = vu;Vpj = vpi) 0.49 P3 = P r(Vu = vu;Vp1 = vpi) 0.03 2 1 2 2 P2 = P r(Vu = vu;Vp1 = vpi) 0.19 P4 = P r(Vu = vu;Vp1 = vpi) 0.28

29 Results for the baseline hazard from unemployment show negative duration dependence, the exit rates decrease with the time spent on UI bene…t. The biggest decrease, however, is noticeable during the …rst half year of unemployment, while after one year the baseline hazards remain rather stable. The baseline hazards to di¤erent types of ALMPs show the same magnitute for both men and women. The hazards remain very low during the …rst six months of the UI bene…t spell, rise thereafter and reach a spike in the third half year of unemployment as they should according to the rules. Concerning the personal characteristics, we …nd a negative e¤ect of age on the exit from unemployment and to ALMPs for men. Being 30-39 years old reduces the hazard rate out of unemployment by 6.4%, belonging to the 40-49 age group has an impact of -16.4%, while the persons over …fty experience a 38.4% reduction in their hazard rate, compared with the reference group - 25-29 years old individuals. The chance to start private job training decreases by age, while 30-39 years old men have the lowest risk of a temporary job in a public sector (the hazard lowers by 46.6%, compared to their 25-29 years old counterparts). Being over thirty, but less than …fty reduces the hazards to education and other ALMPs by one …fth, and those over …fty experience about one third of the hazard rate reduction. For women, however, those in the 30-39 age group have slightly higher exit rate from un- employment (by 6.7%), but the reduction in the hazard for elderly unemployed women is very close to that of men. Like in the case of men we …nd a negative e¤ect of age to transitions to private job training and a big reduction of moving to the temporary public job for 30-39 years old women. However, belonging to the middle age group increases the probability of training. Being single lowers the hazards out of UI - by 20.5% and 12.5% for men and women respec- tively - and make women more likely to be activated in education and other ALMPs. For single men the hazard to private job training reduces. Turning to the immigration status, it is important to distinguish between the immigrants from the developed and less developed countries. While the male immigrants from DCs face a 15.2% reduction in their hazard rate from unemployment (compared to the natives), the exit rate from UI is much lower for immigrants from LDCs. The probability of education increases for immigrants from LDCs and for male immigrants from DCs, while for female immigrants from DCs the exit to other ALMPs is higher than for natives.

30 The unemployment insurance fund membership also plays an important role in explaining the transitions from unemployment and to ALMPs. The members of most UI funds face a lower hazard from unemployment than the reference category - SID+KAD - which are the two main insurance funds for unskilled men and women. The exceptions here are male members of UI fund Construction (12.4% increase in the hazard) and female members of UI fund Clerical (29.5% increase) and UI fund Construction (5.7% increase), while self-employed individuals are found to face the highest risk to be trapped in long term unemployment. Membership of all UI funds reduces the hazards to private and public job training compared to the unskilled workers in the reference UI fund (for both men and women, and by 21.8% - 87%; see Tables 9-10) and in most cases lowers transitions to education and other ALMPs, but to a lower extent. And …nally, concerning past labour market history, we …nd that staying on public income transfers in the past two years lowers the hazard rate out of unemployment for men. Transitions to private job training and other ALMPs are lower for men, while the exit to education increases for women. Experiencing sickness spells during the last year and during the last two years has a di¤erential and complicated impact on the exits from unemployment. 2 1 Concerning unobserved heterogeneity, we …nd that 55% of men (P r(Vu = vu;Vp1 = vpi) + 2 2 P r(Vu = vu;Vp1 = vpi) = 0:55) and 47% of women experience di¢ culties to exit unemployment due to unobserved characteristics, i.e. their hazard rate from UI is reduced by 46.7% (for men) and 53.5% (for women). And 40% of these men and 60% of women (or 22% and 28% of 2 2 2 all unemployed men and women; P r(Vu = vu;Vp1 = vpi; men) = 0:22 and P r(Vu = vu;Vp1 = 2 vpi; women) = 0:28) face a much lower probability of activation in the private sector (see Tables 9-10).

8. Meta Analysis Findings

In the previous sections, we have discussed the performance of Danish ALMPs in di¤erent coun- ties of Denmark. We found that the di¤erent ALMPs had di¤erent in‡uences on unemployed men and women. The decrease in unemployment duration, resulting from activation in private job training programmes, was higher for women, public job training and other ALMPs behaved better for men, while education had the same impact on the unemployment duration for men

31 and women. We also found that the programmes behaved very di¤erently in the di¤erent coun- ties. The most e¤ective programmes - private job training - had an impact on unemployment duration from a 1.87 weeks decrease (for women in Copenhagen and Frederiksberg region) to a 0.86 weeks increase (for women in Ribe). Public sector employment programmes increased unemployment duration by 0.55 - 2.48 weeks, education programmes - by 1.31 - 2.7 weeks, while the impact of other ALMPs varied from -1.16 to 1.7 weeks. An interesting …nding is that in general participation in ALMPs a¤ected women more than men. This should be taken into consideration, since unemployment duration for women in Danish counties is considerably longer than for men. One important aspect is the intensity of programme use (which is de…ned as the fraction of the unemployed individuals, activated in the particular type of ALMPs; individuals, acti- vated in all four ALMPs types, equal to 100%). Tables A.7.1 –A.7.4 in the Appendix present the intensity of di¤erent types of ALMPs among the unemployed men and women in Danish counties, and in the whole country in 1999 –2004. The tables tell us that the education pro- grammes, which have shown the poorest performance in the employability of men and women, tend to be the programmes which are used most for both men and women. More than half of men and women within the country, who participate in programmes, participate in education programmes, while private job training programmes were used the least. On the other hand, the use of private job training increased from 5.8% in 1999 to 9.5% in 2004 for men and from 3.2% to 5.5% for women. There are annual variations in the intensity of programme use. Particular interest here has to be paid to the year 2002, when the popularity of private job training programmes within the country rose from 7.7% to 19.7% for men and from 4.4% to 9.7% for women. The use of other ALMPs went down - from 23% to 11% and 20.2% to 10.6% respectively. The fraction of activation in public job training increased while activation in education/training declined, though not so signi…cantly. In 2003, however, these …gures went in the opposite direction, back towards the 2001 level. The intensity of programme use thus di¤ers among years within the county level. However, most of the counties experienced quite similar annual ‡uctuations. The goal of this chapter is to investigate how the behaviour of the ALMPs in Denmark depends on the regional unemployment level and duration, the composition of the work force

32 (skills), the structure of labour market (sectoral composition), the age of the unemployed, and …nally, on the intensity of the use of the programmes. We employ here a meta analysis, described in Section 5, to the results presented in the last chapter. A number of explanatory factors (see Section 5) are considered. Tables 11-12 present results of the analysis for men (excluding and including year dummies), and tables 13-14 present the same results for women. Signi…cant estimates here are marked in bold. One of the most important …ndings of the analysis, common for men and women, is the existence of highly signi…cant relationships between the expected unemployment duration and the e¤ects of public job training and education programmes. When expected unemployment duration increases by 1 week, participation in public job training programmes prolongs the unemployment duration by 0.085 weeks for men and 0.077 weeks for women. For education these …gures are 0.065 and 0.055 weeks, respectively. It is disappointing that the programmes, which already showed the worst impact, tend to be the most harmful in the counties (and years) with long unemployment durations. This relation is estimated for given levels of the regional unemployment rate, so the interpretation is that regions with low turnover rates in the labour market (an immobile labour market) show worse impacts of programme participation. This result must be investigated further and pursued in several directions, one of which is the optimal program entry time. For men, the fraction of those, older than 50 years, shows a statistically signi…cant correla- tion with the e¤ectiveness of education programmes. Since education programmes are mainly o¤ered to young workers, this could imply that they are more e¤ective when there is less com- petition for the skills they provide. The relationship between intensity of programme use (activation, %) and programme e¤ec- tiveness also needs to be explored further. On the one hand, we would expect an increasing use of a programme to render it less e¤ective, since increasing the intensity of programme use is assumed to lead to diminishing marginal returns. On the other hand, it may be the case that there is an e¤ectiveness gain in the learning sense; as programmes are used more, the administrators learn how to improve their e¤ectiveness; kind of a ’returns to scale’argument. In the results, there seem to be indication of both, but we are considering including both the relative distribution of programmes (as we have now) as well as the absolute number of persons

33 in di¤erent programmes, in order to capture both e¤ects. The learning story would then - presumably - be captured by the latter variable. Private sector employment programmes tend to behave worse (for women) in counties where the fraction of self-employed workers is larger, but education programmes showed the opposite tendency. The implication may be that in regions with many small …rms (many self-employed individuals), there is a larger need for upgrading the skills of the work force (presumably due to many small business start-ups, which require skilled workers to achieve high productivity levels, making them able to compete in the market). At the same time, job-openings are less likely to occur in small …rms, since for a given turnover rate, more persons will be leaving large …rms, rendering private job training programmes less e¤ective in such regions. In the model without year dummies, the increase in activation of women in private job train- ing is found to be related to the decrease in their unemployment duration. These programmes also performed better when the fraction of the low-skilled was bigger, while public job training showed the opposite. When including dummies, these e¤ects are no longer signi…cant. Thus, the signi…cant coe¢ cients tend to be caused by the annual di¤erences in activations and in the fraction of low-skilled that happen to be correlated with the net e¤ect of the programmes, rather than by cross-county variation. The fractions of the individuals, employed in manufac- turing and in the private sector, did not show any signi…cant relations with the net e¤ects of the ALMPs, neither for men nor for women. There are signi…cant annual e¤ects on the e¢ ciency of ALMPs, and here attention has to be paid to the year 2002. As it was mentioned above, the intensity of programme use di¤ered in 2002, when the use of private job training increased quite strongly. Nevertheless, counter intuitively, we found that private job training programmes were extremely e¤ective in 2002.

Based on the …ndings above we can conclude, that the e¤ectiveness of Danish ALMPs to men does not seem to depend neither on the composition of the work force (skills) nor on the labour market (sectoral composition). For women, the sectoral composition does not in‡uence the e¤ect of the ALMPs, but skill composition plays a signi…cant role. The unemployment level does not seem to be an important factor.

34 Table 11. E¤ectiveness of Danish ALMPs for Men

Private job tr. Public job tr. Education Other ALMPs E¤ect t-ratio E¤ect t-ratio E¤ect t-ratio E¤ect t-ratio Activation, % -0.012 -0.95 0.005 0.43 -0.005 -2.29 0.002 0.59 Low-skilled, % -0.002 -0.10 0.030 1.54 0.004 0.52 -0.023 -1.61 Manufacturing, % -0.007 -0.17 -0.047 -1.55 0.005 0.39 0.007 0.33 Private sector, % -0.014 -0.43 0.017 0.66 0.017 1.59 -0.001 -0.04 Self-employed, % 0.032 0.40 0.016 0.24 -0.030 -1.08 0.017 0.37 Unemployment, % -0.005 -0.08 -0.043 -0.79 0.038 1.81 0.013 0.37 >50 years old, % -0.027 -0.81 -0.005 -0.18 -0.038 -3.55 0.010 0.51 Duration, weeks -0.038 -0.88 0.084 2.43 0.067 4.54 0.032 1.27

Table 12. E¤ectiveness of Danish ALMPs for Men (with Year Dummies)

Private job tr. Public job tr. Education Other ALMPs E¤ect t-ratio E¤ect t-ratio E¤ect t-ratio E¤ect t-ratio Activation, % 0.045 2.17 -0.005 -0.29 -0.006 -2.56 -0.001 -0.19 Low-skilled, % -0.029 -0.74 0.049 1.42 -0.012 -0.86 0.012 0.48 Manufacturing, % -0.043 -1.00 -0.057 -1.63 0.017 1.21 -0.015 -0.57 Private sector, % 0.012 0.38 0.015 0.57 0.011 1.04 -0.004 -0.22 Self-employed, % 0.092 1.22 0.022 0.32 -0.038 -1.47 0.015 0.32 Unemployment, % 0.022 0.29 -0.051 -0.75 0.033 1.28 -0.064 -1.34 >50 years old, % -0.053 -1.61 -0.002 -0.06 -0.032 -3.06 -0.005 -0.24 Duration, weeks -0.048 -1.20 0.085 2.43 0.064 4.69 0.030 1.18 Year dummies 2000 0.080 0.38 -0.284 -1.65 -0.152 -2.25 -0.042 -0.34 2001 -0.451 -2.00 -0.014 -0.07 -0.246 -3.06 0.044 0.31 2002 -1.159 -3.09 0.109 0.41 -0.195 -2.04 0.010 0.07 2003 -0.689 -2.29 0.046 0.18 -0.073 -0.72 0.305 1.88 2004 -0.365 -1.00 0.054 0.17 -0.252 -1.87 0.376 1.66

35 Table 13. E¤ectiveness of Danish ALMPs for Women

Private job tr. Public job tr. Education Other ALMPs E¤ect t-ratio E¤ect t-ratio E¤ect t-ratio E¤ect t-ratio Activation, % -0.172 -2.75 -0.006 -0.47 0.002 0.39 0.017 2.37 Low-skilled, % -0.165 -2.14 0.091 2.04 -0.012 -0.52 0.013 0.32 Manufacturing, % 0.185 1.84 -0.019 -0.33 0.029 1.01 -0.020 -0.40 Private sector, % -0.052 -0.81 -0.007 -0.17 -0.006 -0.33 0.026 0.72 Self-employed, % 1.328 2.57 -0.099 -0.32 -0.477 -3.02 0.203 0.70 Unemployment, % -0.056 -0.46 -0.081 -1.10 0.004 0.12 0.187 2.75 >50 years old, % -0.019 -0.23 -0.021 -0.41 0.012 0.49 -0.064 -1.41 Duration, weeks 0.046 0.78 0.077 2.13 0.057 3.20 0.031 0.95

Table 14. E¤ectiveness of Danish ALMPs for Women (with Year Dummies)

Private job tr. Public job tr. Education Other ALMPs E¤ect t-ratio E¤ect t-ratio E¤ect t-ratio E¤ect t-ratio Activation, % -0.001 -0.013 0.012 0.60 -0.003 -0.63 0.015 2.09 Low-skilled, % 0.034 0.258 0.068 0.87 0.006 0.16 2.088 0.38 Manufacturing, % -0.028 -0.213 -0.025 -0.35 0.007 0.23 -0.044 -0.73 Private sector, % 0.019 0.268 0.007 0.16 0.006 0.31 0.040 1.13 Self-employed, % 1.992 3.43 -0.079 -0.23 -0.344 -2.22 0.486 1.64 Unemployment, % -0.277 -1.82 -0.082 -0.93 -0.034 -0.84 0.106 1.36 >50 years old, % -0.161 -1.46 0.006 0.09 0.011 0.36 -0.085 -1.49 Duration, weeks 0.026 0.445 0.071 2.00 0.053 3.51 0.026 0.91 Year dummies 2000 -0.366 -0.73 -0.700 -2.56 -0.365 -3.11 -0.155 -0.68 2001 -0.144 -0.27 -0.723 -2.31 -0.607 -4.23 -0.704 -2.71 2002 -0.981 -1.13 -0.865 -1.87 -0.507 -2.83 -0.623 -2.24 2003 0.405 0.56 -0.724 -1.61 -0.175 -0.86 0.272 0.79 2004 1.030 1.20 -0.605 -1.14 -0.079 -0.31 -0.008 -0.02

36 The most important …nding is that public job training and education programmes tend to behave worse when unemployment duration increases. This is very disappointing since not only the unemployment rate, but also the duration of unemployment is an important measure of welfare (see Borooah, 2002). This suggests that these programmes should be used more actively in regions with higher levels of labour turnover (labour markets characterised by high mobility).

9. Conclusions

In this study we have used the timing-of-events model - developed by Abbring & van den Berg (2003) for identifying treatment e¤ects non-parametrically in a duration model framework - to estimate the impact of ALMPs on the escape rate from unemployment for men and women in 14 Danish counties. We have used fairly new Danish data from administrative registers to estimate the parameters of the econometric model. The e¤ects of ALMPs have been decomposed in two separate e¤ects - a locking-in e¤ect and a post-programme e¤ect - and the model has been estimated separately for each county, and for men and women. Calculations of net impacts on the expected duration of unemployment reveal that only private sector job training reduce unemployment duration. As programme impacts are esti- mated separately by region and year, we subsequently conduct a meta-analysis, relating the net impacts of each programme to a number of variables, describing the local labour market, the composition of the work force, and the characteristics of unemployment and programme use. We found a tendency for programmes to be least e¤ective for those who need them the most; those with longer unemployment durations. This important result has a number of policy im- plications, as the implied trade-o¤ between equity and e¢ ciency in the construction of ALMPs would tend to favour other policies than the two, which are used most intensively.

37 References

[1] Abbring, J. and G. van den Berg (2003). “The Non-Parametric Identi…cation of Treatment E¤ects in Duration Models”, Econometrica, 71, 1491-1517.

[2] Andersen, T. M. and M. Svarer (2006), "Flexicurity - den danske arbetsmarknadsmod- ellen". Ekonomisk Debatt, Vol. 34, 17-29.

[3] Ashenfelter, O., C. Harmon and H. Oosterbeek (1999). "A Review of Estimates of the Schooling/Earnings Relationship, with Tests for Publication Bias", Labour Economics, 6(4):453-70.

[4] Black, D., J. A. Smith, M. C. Berger and B. J. Noel (2003). "Is the Threat of Reem- ployment Services More E¤ective than the Services Themselves?", Amercian Economic Review, Vol. 93, 1313-1327.

[5] Bolvig, I., P. Jensen and M. Rosholm (2003). "The Employment E¤ects of Active Social Policy", IZA Discussion Paper No. 736, Institute for the Study of Labor, Bonn.

[6] Borooah, V. K. (2002). "A Duration-Sensitive Measure of the Unemployment Rate: Theory and Application", ICER Working Paper No 19-2002.

[7] Card, D. and A.B. Krueger (1995). "Time-Series Minimum-Wage Studies: A Meta- analysis", The American Economic Review, Vol. 85(2):238-243.

[8] Graversen, B. K. (2003). "The Impact of Active Labour Market Programmes on Welfare Dependence in Denmark", The Danish National Institute of Social Research, Aarhus, mimeo.

[9] Hammer, B., M. Rosholm and M. Svarer (2004). “A Danish Pro…ling System”, Unpub- lished Manuscript, University of Aarhus.

[10] Kluve, J. and C.M. Schmidt (2002). "Can training and employment subsidies combat European unemployment?", Economic Policy 35, 409-448.

[11] Kluve, J. (2003). "The E¤ectiveness of European Active Labor Market Policy", IZA Discussion Paper No. 2018, Institute for the Study of Labor, Bonn.

38 [12] Munch, J. R. & L. Skipper (2004). "The Consequences of Active Labour Market Programme Participation in Denmark", Unpublished Working Paper.

[13] Phillips, J.M. (1994). "Farmer Education and Farm E¢ ciency", Economic Development and Change 43(1):149-165.

[14] Rosholm, M. and M. Svarer (2004). “Estimating the Threat E¤ect of Active Labour Market Programmes”, Working Paper No. 2004-06, University of Aarhus.

[15] Røed, K. and L. Westlie (2007). "Unemployment Insurance in Welfare States: Soft Constraints and Mild Sanctions", IZA Discussion Paper No. 2877, Institute for the Study of Labor, Bonn.

[16] Van der Sluis, J., C. M. van Praag and W. Vijverberg (2004). "Education and Entre- preneurship in Industrialized Countries: A Meta-analysis", Tinbergen Institute Working Paper No. TI 03-046/3.

[17] Van der Sluis, J., C. M. van Praag and W. Vijverberg (2005). "Entrepreneurship selec- tion and performance: A meta-analysis of the impact of education in less developed countries", World Bank Economic Review, 19 , (2), 225-261.

39 APPENDIX

TABLE A.1. DESCRIPTION OF THE SAMPLE (OBSERVATIONS)

Men Women Copenhagen and Frederiksberg 72363 86303 Copenhagen county 45451 66191 Frederiksborg county 23998 38646 Roskilde county 16713 28928 Western Zelland county 28297 44260 Storstrøms county 29526 42913 Funen county 61298 82768 Southern Jutland county 25424 42637 Ribe county 23698 36695 Vejle county 35227 59830 Ringkøbing county 27082 51998 Århus county 79193 116823 Viborg county 26098 40951 Northern Jutland county 71980 102534

40 TABLE A.2.1. LOCKING-IN EFFECTS OF PRIVATE JOB TRAINING TO MEN (1999-2001)

1999 2000 2001 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •0.382 0.569 •31.7 •0.565 0.382 •43.2 •0.152 0.279 •14.1 Copenhagen county •0.331 0.820 •28.2 •0.029 0.545 •2.9 •0.258 0.345 •22.7 Frederiksborg county • • • •0.811 0.764 •55.6 0.189 0.353 20.8 Roskilde county 0.260 1.350 29.7 •0.458 0.905 •36.8 0.027 0.493 2.7 Western Zelland county 0.445 0.679 56.0 •0.237 0.502 •21.1 0.220 0.392 24.7 Storstrøms county •0.659 0.881 •48.3 •0.336 0.377 •28.5 •0.399 0.339 •32.9 Funen county •0.640 0.972 •47.3 •0.620 0.295 •46.2 •0.094 0.211 •9.0 Southern Jutland county •0.860 0.796 •57.7 •0.354 0.368 •29.8 •0.287 0.296 •24.9 Ribe county 0.224 6.537 25.0 •0.622 0.605 •46.3 •0.362 0.790 •30.4 Vejle county •0.555 0.645 •42.6 0.023 0.435 2.3 •0.123 0.338 •11.6 Ringkøbing county •0.299 0.764 •25.8 •0.259 0.456 •22.8 0.139 0.522 15.0 Århus county •0.880 0.319 •58.5 •0.242 0.224 •21.5 •0.357 0.204 •30.0 Viborg county •0.461 0.547 •36.9 •1.052 0.709 •65.1 •0.885 0.533 •58.7 Northern Jutland county •0.408 0.519 •33.5 •0.279 0.421 •24.3 •0.401 0.240 •33.0

TABLE A.2.2. LOCKING-IN EFFECTS OF PRIVATE JOB TRAINING TO MEN (2002-2004)

2002 2003 2004 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg 0.041 0.211 4.2 •0.281 0.228 •24.5 •0.477 0.159 •37.9 Copenhagen county 0.213 0.304 23.7 •0.001 0.260 •0.1 •0.496 0.214 •39.1 Frederiksborg county 0.339 0.591 40.4 •0.071 0.365 •6.8 •0.188 0.256 •17.2 Roskilde county 0.382 0.478 46.5 0.145 0.436 15.7 •0.547 0.282 •42.2 Western Zelland county •0.194 0.367 •17.6 0.300 0.237 35.0 •0.148 0.292 •13.7 Storstrøms county •0.084 0.275 •8.1 •0.371 0.350 •31.0 •0.303 0.241 •26.1 Funen county •0.184 0.230 •16.8 •0.161 0.195 •14.9 •0.255 0.150 •22.5 Southern Jutland county 0.859 0.302 136.0 •0.078 0.365 •7.5 •0.074 0.349 •7.2 Ribe county •0.373 0.486 •31.1 •0.369 0.399 •30.8 •0.670 0.399 •48.8 Vejle county 0.109 0.283 11.5 0.186 0.304 20.4 •0.031 0.233 •3.0 Ringkøbing county •0.169 0.380 •15.5 0.180 0.343 19.8 •0.454 0.293 •36.5 Århus county 0.000 0.202 0.0 •0.317 0.192 •27.2 •0.210 0.120 •18.9 Viborg county •0.316 0.354 •27.1 •0.219 0.314 •19.6 •0.540 0.302 •41.7 Northern Jutland county •0.338 0.223 •28.7 •0.099 0.211 •9.4 •0.232 0.139 •20.7

41 TABLE A.2.3. LOCKING-IN EFFECTS OF PRIVATE JOB TRAINING TO WOMEN (1999-2001)

1999 2000 2001 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •0.727 1.080 •51.7 0.040 0.662 4.1 •0.296 0.410 •25.6 Copenhagen county •0.460 1.259 •36.9 •0.022 0.398 •2.1 •0.252 0.619 •22.3 Frederiksborg county 0.268 1.360 30.8 •0.664 0.711 •48.5 •0.793 0.654 •54.7 Roskilde county 0.351 2.323 42.1 •0.280 1.102 •24.4 0.093 0.655 9.7 Western Zelland county 0.493 0.952 63.7 •0.495 0.687 •39.0 •0.256 0.583 •22.5 Storstrøms county •0.480 0.645 •38.1 0.726 0.447 106.7 •0.407 0.625 •33.4 Funen county •0.752 0.663 •52.9 •0.708 0.686 •50.7 •0.318 0.390 •27.2 Southern Jutland county 0.217 0.638 24.2 •0.114 0.417 •10.8 •0.109 0.315 •10.4 Ribe county •0.896 1.277 •59.2 •0.762 0.729 •53.3 •0.287 0.457 •25.0 Vejle county •0.555 0.501 •42.6 0.023 0.325 2.3 •0.123 0.233 •11.6 Ringkøbing county •0.287 0.507 •24.9 0.089 0.296 9.3 0.385 0.289 46.9 Århus county •0.805 0.428 •55.3 •0.231 0.266 •20.6 •0.307 0.231 •26.4 Viborg county •1.500 1.113 •77.7 •0.078 0.406 •7.5 •0.155 0.435 •14.4 Northern Jutland county •0.415 0.429 •34.0 •0.466 0.374 •37.2 0.116 0.225 12.3

TABLE A.2.4. LOCKING-IN EFFECTS OF PRIVATE JOB TRAINING TO WOMEN (2002-2004)

2002 2003 2004 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •0.468 0.344 •37.4 0.380 0.259 46.2 •0.369 0.208 •30.8 Copenhagen county 0.493 0.368 63.8 •0.197 0.293 •17.9 •0.235 0.327 •20.9 Frederiksborg county •0.353 0.744 •29.8 •0.353 0.723 •29.8 •0.462 0.433 •37.0 Roskilde county 0.837 0.569 131.0 •0.024 0.633 •2.3 •0.754 0.447 •52.9 Western Zelland county •0.373 0.346 •31.1 •0.379 0.509 •31.6 •0.899 0.416 •59.3 Storstrøms county •0.467 0.489 •37.3 •0.472 0.388 •37.6 •0.630 0.341 •46.8 Funen county 0.055 0.285 5.7 •0.200 0.361 •18.1 •0.122 0.201 •11.5 Southern Jutland county 0.007 0.337 0.7 •0.740 0.926 •52.3 •0.135 0.350 •12.6 Ribe county •0.310 0.557 •26.6 •0.145 0.426 •13.5 •0.654 0.389 •48.0 Vejle county 0.109 0.261 11.5 0.186 0.262 20.4 •0.031 0.229 •3.0 Ringkøbing county 0.415 0.350 51.4 •0.060 0.386 •5.8 0.041 0.380 4.2 Århus county 0.216 0.205 24.1 0.037 0.197 3.8 •0.143 0.158 •13.3 Viborg county 0.039 0.361 4.0 •0.095 0.442 •9.1 •0.400 0.362 •32.9 Northern Jutland county 0.290 0.248 33.7 •0.262 0.300 •23.0 •0.035 0.192 •3.5

42 TABLE A.2.5. LOCKING-IN EFFECTS OF PUBLIC JOB TRAINING TO MEN (1999-2001)

1999 2000 2001 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •0.906 0.274 •59.6 •1.029 0.242 •64.3 •0.939 0.219 •60.9 Copenhagen county •1.116 0.276 •67.2 •0.945 0.350 •61.1 •1.276 0.285 •72.1 Frederiksborg county •1.948 0.717 •85.7 •1.091 0.520 •66.4 •1.476 0.548 •77.1 Roskilde county •0.884 1.840 •58.7 •1.740 0.799 •82.4 •1.080 0.539 •66.0 Western Zelland county •1.747 0.566 •82.6 •1.219 0.554 •70.5 •0.888 0.412 •58.9 Storstrøms county •1.201 0.412 •69.9 •0.836 0.355 •56.6 •0.966 0.257 •61.9 Funen county •2.242 1.079 •89.4 •1.411 0.383 •75.6 •1.629 0.346 •80.4 Southern Jutland county •1.904 0.776 •85.1 •1.389 0.456 •75.1 •0.643 0.383 •47.4 Ribe county •1.190 0.654 •69.6 •1.375 0.701 •74.7 •0.409 0.562 •33.6 Vejle county •1.150 1.004 •68.3 •0.710 0.693 •50.8 •1.034 0.625 •64.4 Ringkøbing county •1.073 1.138 •65.8 •0.751 0.743 •52.8 •0.992 0.255 •62.9 Århus county •1.220 0.405 •70.5 •0.815 0.268 •55.7 •1.007 0.245 •63.5 Viborg county • • • •0.963 0.438 •61.8 •1.543 0.495 •78.6 Northern Jutland county •1.077 0.610 •65.9 •1.086 0.376 •66.3 •0.823 0.234 •56.1

TABLE A.2.6. LOCKING-IN EFFECTS OF PUBLIC JOB TRAINING TO MEN (2002-2004)

2002 2003 2004 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •1.026 0.170 •64.1 •0.898 0.150 •59.3 •0.903 0.160 •59.5 Copenhagen county •0.767 0.215 •53.6 •0.776 0.241 •54.0 •0.705 0.198 •50.6 Frederiksborg county •0.966 0.389 •61.9 •0.354 0.252 •29.8 •0.911 0.434 •59.8 Roskilde county •0.953 0.535 •61.4 •1.822 0.835 •83.8 •1.007 0.414 •63.5 Western Zelland county •0.749 0.377 •52.7 •1.568 0.592 •79.2 •0.958 0.317 •61.6 Storstrøms county •0.642 0.266 •47.4 •1.172 0.339 •69.0 •0.747 0.221 •52.6 Funen county •1.175 0.297 •69.1 •1.255 0.261 •71.5 •0.732 0.201 •51.9 Southern Jutland county •0.507 0.384 •39.7 •0.667 0.349 •48.7 •0.819 0.364 •55.9 Ribe county •0.710 0.469 •50.8 •0.820 0.480 •56.0 •0.430 0.367 •34.9 Vejle county •0.371 0.445 •31.0 •1.034 0.520 •64.5 •0.697 0.393 •50.2 Ringkøbing county •1.454 1.007 •76.6 •0.543 0.455 •41.9 •1.028 0.527 •64.2 Århus county •0.559 0.204 •42.8 •0.674 0.198 •49.0 •0.427 0.181 •34.8 Viborg county •0.604 0.573 •45.4 •1.155 0.735 •68.5 •1.037 0.444 •64.6 Northern Jutland county •0.590 0.220 •44.6 •0.851 0.251 •57.3 •0.172 0.206 •15.8

43 TABLE A.2.7. LOCKING-IN EFFECTS OF PUBLIC JOB TRAINING TO WOMEN (1999-2001)

1999 2000 2001 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •1.881 0.480 •84.8 •0.936 0.247 •60.8 •0.718 0.203 •51.2 Copenhagen county •1.239 0.356 •71.0 •1.520 0.280 •78.1 •0.574 0.197 •43.7 Frederiksborg county •1.886 0.441 •84.8 •0.846 0.396 •57.1 •1.269 0.395 •71.9 Roskilde county •1.241 0.527 •71.1 •0.584 0.346 •44.2 •0.901 0.338 •59.4 Western Zelland county •1.518 0.416 •78.1 •0.908 0.269 •59.7 •1.241 0.300 •71.1 Storstrøms county •0.998 0.276 •63.1 •0.555 0.283 •42.6 •0.764 0.230 •53.4 Funen county •1.427 0.329 •76.0 •1.677 0.306 •81.3 •0.742 0.180 •52.4 Southern Jutland county •1.538 0.548 •78.5 •1.115 0.401 •67.2 •1.263 0.313 •71.7 Ribe county •1.882 0.478 •84.8 •1.361 0.360 •74.4 •0.837 0.243 •56.7 Vejle county •1.150 0.427 •68.3 •0.710 0.297 •50.8 •1.034 0.325 •64.4 Ringkøbing county •1.428 0.384 •76.0 •0.925 0.284 •60.4 •0.585 0.237 •44.3 Århus county •1.240 0.330 •71.1 •0.923 0.260 •60.3 •0.623 0.193 •46.4 Viborg county •1.855 0.528 •84.4 •1.418 0.424 •75.8 •0.714 0.236 •51.0 Northern Jutland county •1.682 0.384 •81.4 •1.250 0.245 •71.4 •0.873 0.199 •58.2

TABLE A.2.8. LOCKING-IN EFFECTS OF PUBLIC JOB TRAINING TO WOMEN (2002-2004)

2002 2003 2004 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •1.119 0.195 •67.3 •0.612 0.161 •45.8 •0.685 0.147 •49.6 Copenhagen county •0.699 0.186 •50.3 •0.816 0.206 •55.8 •0.645 0.171 •47.5 Frederiksborg county •0.719 0.278 •51.3 •0.242 0.277 •21.5 •0.685 0.321 •49.6 Roskilde county •0.535 0.322 •41.5 •0.932 0.374 •60.6 •0.481 0.254 •38.2 Western Zelland county •0.858 0.254 •57.6 •0.736 0.308 •52.1 •0.781 0.246 •54.2 Storstrøms county •0.893 0.234 •59.1 •0.634 0.212 •46.9 •0.914 0.196 •59.9 Funen county •1.769 0.240 •83.0 •1.769 0.244 •83.0 •1.105 0.172 •66.9 Southern Jutland county •0.635 0.238 •47.0 •0.823 0.252 •56.1 •0.792 0.228 •54.7 Ribe county •0.753 0.273 •52.9 •0.553 0.293 •42.5 •1.109 0.311 •67.0 Vejle county •0.371 0.247 •31.0 •1.034 0.291 •64.5 •0.697 0.231 •50.2 Ringkøbing county •0.571 0.237 •43.5 •1.198 0.321 •69.8 •0.549 0.249 •42.3 Århus county 0.323 0.184 38.1 •0.636 0.147 •47.1 •0.495 0.131 •39.0 Viborg county •0.761 0.277 •53.3 •0.438 0.275 •35.5 •0.774 0.256 •53.9 Northern Jutland county •0.725 0.192 •51.6 •0.508 0.175 •39.8 •0.434 0.153 •35.2

44 TABLE A.2.9. LOCKING-IN EFFECTS OF EDUCATION TO MEN (1999-2001)

1999 2000 2001 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •0.780 0.124 •54.1 •0.856 0.113 •57.5 •0.856 0.087 •57.5 Copenhagen county •1.005 0.190 •63.4 •0.873 0.154 •58.2 •0.703 0.124 •50.5 Frederiksborg county •1.313 0.260 •73.1 •0.782 0.180 •54.2 •0.771 0.168 •53.8 Roskilde county •0.982 0.377 •62.6 •0.714 0.258 •51.0 •0.563 0.223 •43.0 Western Zelland county •0.592 0.239 •44.7 •0.710 0.203 •50.9 •0.387 0.146 •32.1 Storstrøms county •1.635 0.334 •80.5 •0.840 0.211 •56.8 •1.042 0.232 •64.7 Funen county •1.194 0.195 •69.7 •0.757 0.137 •53.1 •1.009 0.126 •63.6 Southern Jutland county •0.949 0.340 •61.3 •1.521 0.268 •78.1 •0.762 0.170 •53.3 Ribe county •0.866 0.418 •57.9 •0.963 0.324 •61.8 •1.249 0.230 •71.3 Vejle county •1.390 0.282 •75.1 •1.102 0.187 •66.8 •1.017 0.171 •63.8 Ringkøbing county •1.268 0.267 •71.9 •1.454 0.287 •76.6 •1.265 0.265 •71.8 Århus county •1.226 0.148 •70.7 •1.169 0.115 •68.9 •1.112 0.094 •67.1 Viborg county •1.609 0.373 •80.0 •1.145 0.454 •68.2 •1.548 0.301 •78.7 Northern Jutland county •1.273 0.174 •72.0 •1.273 0.154 •72.0 •0.791 0.116 •54.7

TABLE A.2.10. LOCKING-IN EFFECTS OF EDUCATION TO MEN (2002-2004)

2002 2003 2004 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •0.881 0.096 •58.6 •0.940 0.112 •60.9 •0.859 0.101 •57.6 Copenhagen county •0.743 0.135 •52.4 •0.677 0.136 •49.2 •0.747 0.138 •52.6 Frederiksborg county •1.299 0.235 •72.7 •0.146 0.202 •13.6 •0.324 0.175 •27.7 Roskilde county •0.563 0.213 •43.0 •0.469 0.192 •37.4 •0.807 0.225 •55.4 Western Zelland county •0.488 0.139 •38.6 •0.628 0.173 •46.6 •0.791 0.235 •54.7 Storstrøms county •0.934 0.220 •60.7 •0.768 0.319 •53.6 •0.745 0.215 •52.5 Funen county •0.975 0.103 •62.3 •0.697 0.106 •50.2 •0.757 0.108 •53.1 Southern Jutland county •0.811 0.218 •55.6 •0.896 0.254 •59.2 •0.572 0.214 •43.6 Ribe county •0.523 0.182 •40.7 •0.859 0.251 •57.6 •0.555 0.174 •42.6 Vejle county •1.057 0.191 •65.3 •1.359 0.226 •74.3 •0.971 0.232 •62.1 Ringkøbing county •1.021 0.238 •64.0 •1.248 0.261 •71.3 •1.006 0.191 •63.4 Århus county •1.107 0.129 •66.9 •0.862 0.109 •57.8 •1.001 0.114 •63.3 Viborg county •0.703 0.196 •50.5 •0.887 0.286 •58.8 •0.810 0.222 •55.5 Northern Jutland county •1.438 0.125 •76.3 •1.015 0.129 •63.8 •0.855 0.126 •57.5

45 TABLE A.2.11. LOCKING-IN EFFECTS OF EDUCATION TO WOMEN (1999-2001)

1999 2000 2001 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •1.177 0.133 •69.2 •1.067 0.113 •65.6 •1.032 0.104 •64.4 Copenhagen county •1.053 0.149 •65.1 •1.257 0.141 •71.6 •0.970 0.113 •62.1 Frederiksborg county •1.388 0.176 •75.0 •1.252 0.150 •71.4 •1.252 0.155 •71.4 Roskilde county •1.599 0.261 •79.8 •1.161 0.211 •68.7 •0.825 0.221 •56.2 Western Zelland county •1.369 0.219 •74.6 •1.160 0.180 •68.7 •0.795 0.134 •54.8 Storstrøms county •1.507 0.233 •77.8 •1.155 0.164 •68.5 •0.848 0.160 •57.2 Funen county •1.577 0.170 •79.3 •1.192 0.128 •69.6 •1.071 0.100 •65.7 Southern Jutland county •1.434 0.198 •76.2 •1.370 0.188 •74.6 •1.178 0.143 •69.2 Ribe county •1.733 0.338 •82.3 •1.091 0.229 •66.4 •1.156 0.176 •68.5 Vejle county •1.390 0.169 •75.1 •1.102 0.153 •66.8 •1.017 0.135 •63.8 Ringkøbing county •1.472 0.175 •77.1 •1.146 0.148 •68.2 •1.197 0.149 •69.8 Århus county •1.261 0.125 •71.7 •1.228 0.102 •70.7 •1.286 0.088 •72.4 Viborg county •1.821 0.281 •83.8 •1.164 0.233 •68.8 •0.843 0.174 •57.0 Northern Jutland county •1.472 0.136 •77.1 •0.919 0.106 •60.1 •1.181 0.103 •69.3

TABLE A.2.12. LOCKING-IN EFFECTS OF EDUCATION TO WOMEN (2002-2004)

2002 2003 2004 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •1.145 0.108 •68.2 •1.094 0.107 •66.5 •1.137 0.111 •67.9 Copenhagen county •0.952 0.107 •61.4 •0.992 0.133 •62.9 •1.224 0.145 •70.6 Frederiksborg county •1.207 0.195 •70.1 •0.947 0.213 •61.2 •0.976 0.232 •62.3 Roskilde county •1.067 0.220 •65.6 •0.907 0.184 •59.6 •1.036 0.217 •64.5 Western Zelland county •0.897 0.130 •59.2 •1.006 0.188 •63.4 •1.068 0.245 •65.6 Storstrøms county •1.091 0.163 •66.4 •0.939 0.224 •60.9 •1.592 0.241 •79.7 Funen county •1.097 0.092 •66.6 •0.923 0.102 •60.3 •0.834 0.109 •56.6 Southern Jutland county •1.234 0.164 •70.9 •1.372 0.222 •74.6 •1.257 0.210 •71.5 Ribe county •1.129 0.158 •67.7 •1.058 0.214 •65.3 •1.583 0.300 •79.5 Vejle county •1.057 0.150 •65.3 •1.359 0.210 •74.3 •0.971 0.181 •62.1 Ringkøbing county •0.926 0.149 •60.4 •1.109 0.185 •67.0 •1.494 0.191 •77.6 Århus county •1.093 0.096 •66.5 •1.026 0.105 •64.1 •1.137 0.108 •67.9 Viborg county •1.014 0.159 •63.7 •1.154 0.216 •68.5 •1.342 0.221 •73.9 Northern Jutland county •1.289 0.093 •72.4 •1.195 0.107 •69.7 •1.402 0.148 •75.4

46 TABLE A.2.13. LOCKING-IN EFFECTS OF OTHER ALMPSs TO MEN (1999-2001)

1999 2000 2001 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •0.718 0.269 •51.2 •0.506 0.289 •39.7 •0.273 0.152 •23.9 Copenhagen county •0.388 0.277 •32.2 •0.389 0.392 •32.2 •0.581 0.213 •44.0 Frederiksborg county •0.219 0.596 •19.7 •1.471 0.636 •77.0 •0.859 0.352 •57.6 Roskilde county •0.503 0.567 •39.5 •0.145 0.973 •13.5 •0.068 0.293 •6.6 Western Zelland county 0.031 0.453 3.1 0.469 0.486 59.9 •0.589 0.333 •44.5 Storstrøms county •1.015 0.537 •63.8 •0.232 0.279 •20.7 •0.108 0.297 •10.2 Funen county 0.362 0.692 43.6 0.207 0.389 23.0 •0.368 0.192 •30.8 Southern Jutland county •1.053 1.138 •65.1 0.476 0.517 61.0 •0.502 0.431 •39.4 Ribe county •0.697 0.539 •50.2 •0.302 0.525 •26.1 0.335 0.437 39.8 Vejle county •1.038 0.329 •64.6 •0.312 0.302 •26.8 •0.240 0.208 •21.3 Ringkøbing county •1.227 0.603 •70.7 •0.503 0.595 •39.5 •0.266 0.255 •23.3 Århus county •0.258 0.274 •22.7 0.089 0.219 9.3 •0.256 0.188 •22.6 Viborg county •0.280 0.393 •24.4 •0.444 0.441 •35.8 •0.004 0.240 •0.4 Northern Jutland county 0.135 0.402 14.4 0.072 0.554 7.5 •0.799 0.502 •55.0

TABLE A.2.14. LOCKING-IN EFFECTS OF OTHER ALMPSs TO MEN (2002-2004)

2002 2003 2004 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •0.093 0.066 •8.9 •0.111 0.066 •10.5 •0.320 0.173 •27.4 Copenhagen county •0.221 0.149 •19.8 •0.287 0.135 •24.9 •0.287 0.178 •24.9 Frederiksborg county •0.970 0.348 •62.1 •0.206 0.536 •18.6 •0.717 0.342 •51.2 Roskilde county •0.168 0.286 •15.4 •0.715 0.799 •51.1 •0.768 0.595 •53.6 Western Zelland county •0.504 0.304 •39.6 •0.110 0.240 •10.4 •0.162 0.169 •15.0 Storstrøms county •0.297 0.229 •25.7 •0.286 0.191 •24.9 •0.110 0.143 •10.4 Funen county •0.256 0.179 •22.6 •0.754 0.563 •52.9 •0.671 0.316 •48.9 Southern Jutland county •0.349 0.330 •29.5 0.586 0.447 79.7 •0.026 0.289 •2.6 Ribe county 0.333 0.397 39.5 0.000 0.000 0.0 •0.949 0.565 •61.3 Vejle county •0.066 0.169 •6.4 •0.656 0.243 •48.1 •0.534 0.180 •41.4 Ringkøbing county •0.479 0.243 •38.1 •0.566 0.341 •43.2 0.098 0.290 10.3 Århus county 0.022 0.145 2.2 •0.360 0.267 •30.2 0.075 0.171 7.8 Viborg county •0.382 0.322 •31.7 0.008 0.530 0.8 •0.902 0.556 •59.4 Northern Jutland county •0.826 0.425 •56.2 •0.707 0.417 •50.7 •0.174 0.177 •16.0

47 TABLE A.2.15. LOCKING-IN EFFECTS OF OTHER ALMPSs TO WOMEN (1999-2001)

1999 2000 2001 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •0.825 0.321 •56.2 •0.792 0.366 •54.7 •0.341 0.192 •28.9 Copenhagen county •0.896 0.384 •59.2 •0.444 0.454 •35.9 •0.145 0.200 •13.5 Frederiksborg county •0.304 0.500 •26.2 •0.304 0.357 •26.2 •0.174 0.245 •16.0 Roskilde county •0.796 0.455 •54.9 •0.262 0.901 •23.0 0.380 0.297 46.2 Western Zelland county 0.017 0.467 1.7 •1.440 0.700 •76.3 0.047 0.252 4.8 Storstrøms county •0.444 0.348 •35.8 •0.314 0.284 •27.0 0.190 0.292 21.0 Funen county •1.435 1.007 •76.2 •0.200 0.389 •18.1 •1.323 0.298 •73.4 Southern Jutland county •0.324 0.745 •27.7 •0.833 0.500 •56.5 •0.225 0.264 •20.1 Ribe county •0.745 0.410 •52.5 •1.228 0.480 •70.7 •1.237 0.592 •71.0 Vejle county •1.038 0.281 •64.6 •0.312 0.254 •26.8 •0.240 0.150 •21.3 Ringkøbing county •1.011 0.288 •63.6 •1.354 0.469 •74.2 •0.840 0.234 •56.8 Århus county •0.664 0.293 •48.5 •0.369 0.240 •30.8 •0.424 0.197 •34.6 Viborg county •1.435 0.435 •76.2 •0.934 0.540 •60.7 •0.227 0.266 •20.3 Northern Jutland county •0.800 0.575 •55.0 •1.110 0.417 •67.1 •1.276 0.479 •72.1

TABLE A.2.16. LOCKING-IN EFFECTS OF OTHER ALMPSs TO WOMEN (2002-2004)

2002 2003 2004 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •0.229 0.130 •20.5 •0.776 0.145 •54.0 •0.666 0.204 •48.6 Copenhagen county •0.227 0.173 •20.3 •0.722 0.167 •51.4 •0.460 0.215 •36.9 Frederiksborg county •0.375 0.245 •31.3 •1.275 0.695 •72.1 •1.086 0.337 •66.2 Roskilde county •0.143 0.341 •13.4 •1.112 1.167 •67.1 •0.009 0.574 •0.9 Western Zelland county •0.219 0.309 •19.7 •0.232 0.276 •20.7 •0.355 0.179 •29.9 Storstrøms county •0.860 0.287 •57.7 •0.184 0.185 •16.8 •0.611 0.203 •45.7 Funen county •0.506 0.178 •39.7 0.153 0.456 16.6 •0.138 0.310 •12.9 Southern Jutland county •0.923 0.341 •60.2 0.328 0.344 38.8 0.232 0.278 26.1 Ribe county 0.037 0.421 3.8 •0.102 0.628 •9.7 •1.368 0.744 •74.5 Vejle county •0.066 0.148 •6.4 •0.656 0.194 •48.1 •0.534 0.164 •41.4 Ringkøbing county •0.611 0.150 •45.7 •0.696 0.256 •50.1 •0.515 0.294 •40.3 Århus county •0.440 0.166 •35.6 •0.222 0.261 •19.9 •0.287 0.084 •24.9 Viborg county •1.066 0.382 •65.5 •0.389 0.549 •32.2 •0.432 0.395 •35.1 Northern Jutland county •0.235 0.289 •20.9 •0.972 0.352 •62.2 •0.132 0.184 •12.3

48 TABLE A.3.1. POST-PROGRAMME EFFECTS OF PRIVATE JOB TRAINING TO MEN

(1999-2001)

1999 2000 2001 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg 0.327 0.257 38.7 0.284 0.247 32.8 0.433 0.222 54.1 Copenhagen county 0.336 0.343 40.0 0.325 0.260 38.4 •0.201 0.303 •18.2 Frederiksborg county 0.621 0.365 86.1 0.507 0.353 66.0 0.517 0.360 67.6 Roskilde county 0.678 0.428 96.9 •0.228 0.288 •20.4 0.757 0.315 113.1 Western Zelland county 0.014 0.435 1.4 0.188 0.355 20.7 0.704 0.324 102.1 Storstrøms county 0.084 0.331 8.8 0.615 0.199 84.9 0.703 0.231 102.0 Funen county 0.568 0.218 76.5 0.261 0.227 29.8 0.474 0.147 60.6 Southern Jutland county 0.321 0.366 37.9 0.125 0.337 13.3 0.321 0.182 37.8 Ribe county 1.435 1.459 320.1 0.585 0.340 79.5 0.907 0.360 147.6 Vejle county 0.234 0.387 26.3 0.391 0.193 47.8 0.609 0.268 83.9 Ringkøbing county 0.726 0.309 106.7 0.422 0.269 52.5 0.889 0.346 143.1 Århus county 0.838 0.170 131.2 0.376 0.176 45.7 0.658 0.158 93.1 Viborg county 0.228 0.297 25.6 •0.078 0.337 •7.5 0.416 0.394 51.6 Northern Jutland county 0.528 0.289 69.5 0.760 0.162 113.8 0.825 0.184 128.2

TABLE A.3.2. POST-PROGRAMME EFFECTS OF PRIVATE JOB TRAINING TO MEN

(2002-2004)

2002 2003 2004 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg 0.356 0.205 42.7 0.793 0.168 120.9 0.253 0.187 28.7 Copenhagen county 0.382 0.281 46.5 0.494 0.240 63.9 0.443 0.187 55.7 Frederiksborg county 0.411 0.302 50.8 •0.165 0.418 •15.2 0.578 0.237 78.2 Roskilde county 0.479 0.378 61.4 0.477 0.406 61.2 0.204 0.323 22.6 Western Zelland county •0.131 0.403 •12.3 0.360 0.206 43.3 0.461 0.226 58.5 Storstrøms county 0.787 0.226 119.6 •0.050 0.270 •4.9 0.333 0.195 39.5 Funen county 0.507 0.184 66.1 0.507 0.143 66.1 0.458 0.104 58.1 Southern Jutland county 0.122 0.232 13.0 0.453 0.333 57.3 0.230 0.219 25.8 Ribe county 0.239 0.365 27.0 0.498 0.247 64.6 0.549 0.197 73.1 Vejle county 0.739 0.196 109.3 0.634 0.197 88.5 0.084 0.140 8.8 Ringkøbing county 0.137 0.310 14.7 0.488 0.284 62.9 0.554 0.186 74.0 Århus county 0.637 0.156 89.1 0.522 0.159 68.5 0.612 0.098 84.5 Viborg county 0.451 0.360 57.0 0.281 0.248 32.4 0.595 0.234 81.2 Northern Jutland county 0.827 0.148 128.7 0.459 0.161 58.3 0.437 0.116 54.9

49 TABLE A.3.3. POST-PROGRAMME EFFECTS OF PRIVATE JOB TRAINING TO WOMEN

(1999-2001)

1999 2000 2001 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg 0.799 0.396 122.3 0.239 0.361 27.0 0.439 0.343 55.1 Copenhagen county 0.028 0.367 2.8 0.143 0.295 15.3 0.074 0.435 7.7 Frederiksborg county •0.425 0.579 •34.6 0.258 0.340 29.4 0.751 0.331 112.0 Roskilde county 0.288 0.607 33.3 0.343 0.850 40.9 0.273 0.575 31.4 Western Zelland county 0.101 0.490 10.6 0.143 0.250 15.3 •0.166 0.504 •15.3 Storstrøms county 0.458 0.269 58.1 •0.345 0.317 •29.2 0.633 0.299 88.2 Funen county 0.725 0.341 106.5 0.706 0.245 102.7 0.706 0.263 102.7 Southern Jutland county 0.005 0.315 0.5 0.414 0.242 51.3 0.323 0.255 38.1 Ribe county 0.673 0.389 95.9 0.145 0.670 15.6 •0.202 0.378 •18.3 Vejle county 0.234 0.255 26.3 0.391 0.286 47.8 0.609 0.212 83.9 Ringkøbing county 0.374 0.209 45.4 •0.027 0.228 •2.6 0.444 0.211 55.8 Århus county 0.521 0.188 68.4 0.772 0.161 116.5 0.564 0.163 75.8 Viborg county •0.063 0.373 •6.1 0.433 0.360 54.2 0.032 0.284 3.2 Northern Jutland county 0.550 0.232 73.2 0.577 0.173 78.1 0.649 0.190 91.4

TABLE A.3.4. POST-PROGRAMME EFFECTS OF PRIVATE JOB TRAINING TO WOMEN

(2002-2004)

2002 2003 2004 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg 0.483 0.264 62.0 0.468 0.258 59.7 0.919 0.209 150.6 Copenhagen county 0.560 0.278 75.1 0.482 0.286 62.0 0.400 0.287 49.2 Frederiksborg county 0.224 0.407 25.1 0.122 0.478 13.0 0.483 0.351 62.0 Roskilde county 0.837 0.719 131.0 0.034 0.421 3.5 0.445 0.447 56.0 Western Zelland county 0.168 0.346 18.3 0.054 0.259 5.6 0.048 0.371 4.9 Storstrøms county 0.499 0.389 64.7 0.348 0.293 41.6 0.726 0.204 106.7 Funen county 0.538 0.196 71.3 0.522 0.182 68.5 0.572 0.149 77.2 Southern Jutland county 0.683 0.250 98.1 0.254 0.367 28.9 0.647 0.259 91.0 Ribe county •0.046 0.325 •4.5 0.103 0.285 10.8 0.383 0.251 46.6 Vejle county 0.739 0.166 109.3 0.634 0.192 88.5 0.084 0.175 8.8 Ringkøbing county 0.270 0.208 31.0 0.329 0.379 39.0 0.452 0.210 57.2 Århus county 0.413 0.158 51.2 0.536 0.141 71.0 0.607 0.128 83.5 Viborg county 0.649 0.305 91.4 0.321 0.279 37.9 0.228 0.251 25.6 Northern Jutland county 0.797 0.183 121.9 0.668 0.173 95.0 0.317 0.167 37.4

50 TABLE A.3.5. POST-PROGRAMME EFFECTS OF PUBLIC JOB TRAINING TO MEN

(1999-2001)

1999 2000 2001 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •0.008 0.181 •0.8 0.086 0.178 9.0 0.116 0.199 12.3 Copenhagen county •0.079 0.242 •7.6 •0.335 0.272 •28.5 •0.298 0.323 •25.8 Frederiksborg county 0.209 0.360 23.3 0.115 0.393 12.1 0.186 0.390 20.4 Roskilde county •0.144 0.668 •13.4 0.578 0.603 78.2 0.116 0.337 12.3 Western Zelland county 0.035 0.348 3.6 •0.335 0.303 •28.4 0.018 0.357 1.8 Storstrøms county •0.157 0.320 •14.5 •0.009 0.248 •0.9 0.152 0.249 16.4 Funen county 0.341 0.280 40.6 0.151 0.204 16.3 •0.240 0.233 •21.3 Southern Jutland county •0.165 0.383 •15.2 0.557 0.356 74.5 0.257 0.374 29.3 Ribe county •0.594 0.608 •44.8 0.463 0.561 58.8 •0.749 0.735 •52.7 Vejle county 0.008 0.398 0.8 0.391 0.508 47.8 0.153 0.866 16.5 Ringkøbing county 0.282 0.489 32.6 0.419 0.531 52.0 •0.323 0.663 •27.6 Århus county •0.009 0.247 •0.9 •0.050 0.179 •4.9 0.172 0.158 18.8 Viborg county •0.206 0.920 •18.6 •0.206 0.434 •18.6 •0.204 0.495 •18.5 Northern Jutland county 0.158 0.248 17.1 •0.122 0.256 •11.5 0.390 0.196 47.6

TABLE A.3.6. POST-PROGRAMME EFFECTS OF PUBLIC JOB TRAINING TO MEN

(2002-2004)

2002 2003 2004 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •0.035 0.161 •3.4 0.067 0.136 6.9 •0.048 0.160 •4.7 Copenhagen county 0.180 0.187 19.7 •0.114 0.214 •10.8 •0.256 0.175 •22.6 Frederiksborg county •0.937 0.568 •60.8 0.252 0.252 28.6 0.066 0.321 6.8 Roskilde county •1.213 0.891 •70.3 •0.064 0.482 •6.2 0.505 0.282 65.7 Western Zelland county 0.227 0.286 25.4 •0.174 0.256 •16.0 •0.049 0.254 •4.8 Storstrøms county •0.366 0.242 •30.6 •0.659 0.331 •48.3 •0.593 0.215 •44.7 Funen county •0.223 0.309 •20.0 •0.311 0.208 •26.7 •0.225 0.178 •20.1 Southern Jutland county •0.574 0.400 •43.7 •0.102 0.295 •9.7 0.135 0.364 14.4 Ribe county •0.151 0.352 •14.0 •0.117 0.299 •11.1 0.129 0.307 13.8 Vejle county •0.040 0.454 •3.9 0.061 0.425 6.3 •0.265 0.381 •23.3 Ringkøbing county •0.342 1.020 •29.0 •0.958 0.822 •61.6 •0.383 0.419 •31.8 Århus county 0.080 0.173 8.3 •0.130 0.164 •12.2 0.080 0.132 8.3 Viborg county •0.184 0.404 •16.8 0.029 0.403 2.9 •0.004 0.416 •0.4 Northern Jutland county •0.146 0.208 •13.6 0.052 0.212 5.4 •0.172 0.142 •15.8

51 TABLE A.3.7. POST-PROGRAMME EFFECTS OF PUBLIC JOB TRAINING TO WOMEN

(1999-2001)

1999 2000 2001 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •0.032 0.232 •3.1 0.265 0.160 30.3 0.114 0.192 12.1 Copenhagen county •0.078 0.213 •7.5 •1.520 0.186 •78.1 •0.149 0.176 •13.8 Frederiksborg county 0.149 0.237 16.1 •0.099 0.212 •9.5 •0.297 0.246 •25.7 Roskilde county 0.008 0.258 0.8 0.112 0.218 11.9 0.052 0.270 5.3 Western Zelland county 0.528 0.234 69.5 0.026 0.215 2.6 0.153 0.192 16.5 Storstrøms county 0.111 0.221 11.7 0.053 0.156 5.4 0.230 0.186 25.8 Funen county 0.147 0.267 15.8 0.246 0.141 27.9 0.041 0.137 4.2 Southern Jutland county 0.060 0.293 6.2 0.044 0.198 4.5 •0.365 0.269 •30.6 Ribe county 0.142 0.227 15.2 •0.017 0.184 •1.7 0.136 0.184 14.5 Vejle county 0.008 0.274 0.8 0.391 0.157 47.8 0.153 0.185 16.5 Ringkøbing county 0.148 0.235 15.9 •0.245 0.215 •21.7 0.295 0.157 34.3 Århus county •0.094 0.137 •8.9 0.219 0.127 24.4 0.311 0.130 36.5 Viborg county 0.088 0.253 9.2 •0.134 0.295 •12.6 0.055 0.154 5.7 Northern Jutland county 0.141 0.176 15.1 0.055 0.158 5.7 0.104 0.139 10.9

TABLE A.3.8. POST-PROGRAMME EFFECTS OF PUBLIC JOB TRAINING TO WOMEN

(2002-2004)

2002 2003 2004 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg 0.224 0.137 25.1 0.337 0.138 40.0 0.080 0.129 8.3 Copenhagen county 0.083 0.153 8.6 •0.188 0.198 •17.1 0.213 0.130 23.8 Frederiksborg county 0.190 0.267 21.0 0.058 0.241 6.0 •0.051 0.265 •5.0 Roskilde county •0.168 0.222 •15.4 •0.330 0.282 •28.1 •0.405 0.252 •33.3 Western Zelland county •0.062 0.179 •6.0 •0.176 0.246 •16.1 •0.223 0.217 •20.0 Storstrøms county •0.186 0.181 •16.9 0.065 0.193 6.7 •0.163 0.180 •15.0 Funen county •0.010 0.138 •1.0 •0.004 0.244 •0.4 •0.068 0.172 •6.6 Southern Jutland county •0.262 0.217 •23.1 •0.823 0.165 •56.1 •0.051 0.154 •4.9 Ribe county •0.119 0.242 •11.2 •0.062 0.231 •6.0 •0.237 0.235 •21.1 Vejle county •0.040 0.200 •3.9 •0.040 0.180 •3.9 •0.265 0.176 •23.3 Ringkøbing county •0.167 0.226 •15.3 0.323 0.168 38.1 0.206 0.159 22.8 Århus county 0.323 0.119 38.1 •0.168 0.122 •15.4 0.073 0.095 7.6 Viborg county 0.035 0.202 3.5 •0.031 0.210 •3.1 0.261 0.162 29.8 Northern Jutland county •0.007 0.143 •0.7 •0.114 0.139 •10.8 0.117 0.114 12.4

52 TABLE A.3.9. POST-PROGRAMME EFFECTS OF EDUCATION TO MEN (1999-2001)

1999 2000 2001 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •0.138 0.081 •12.9 •0.078 0.072 •7.5 0.137 0.065 14.7 Copenhagen county 0.059 0.114 6.1 0.033 0.096 3.3 •0.046 0.089 •4.5 Frederiksborg county 0.133 0.149 14.2 •0.007 0.120 •0.7 0.180 0.124 19.7 Roskilde county 0.218 0.207 24.4 0.171 0.181 18.6 •0.419 0.247 •34.3 Western Zelland county •0.053 0.146 •5.1 0.102 0.128 10.8 0.162 0.121 17.6 Storstrøms county •0.062 0.152 •6.0 0.126 0.122 13.4 0.156 0.125 16.9 Funen county •0.148 0.100 •13.8 0.051 0.089 5.3 0.079 0.081 8.2 Southern Jutland county •0.005 0.192 •0.5 0.207 0.132 22.9 0.086 0.126 9.0 Ribe county •0.148 0.208 •13.8 0.104 0.189 10.9 0.129 0.141 13.7 Vejle county •0.047 0.142 •4.6 •0.132 0.116 •12.4 •0.092 0.118 •8.8 Ringkøbing county •0.143 0.163 •13.3 0.103 0.152 10.9 0.068 0.138 7.0 Århus county •0.060 0.094 •5.8 0.063 0.065 6.5 0.041 0.061 4.2 Viborg county 0.002 0.185 0.2 0.241 0.188 27.3 •1.548 0.177 •78.7 Northern Jutland county 0.034 0.102 3.4 0.006 0.087 0.6 0.231 0.081 26.0

TABLE A.3.10. POST-PROGRAMME EFFECTS OF EDUCATION TO MEN (2002-2004)

2002 2003 2004 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •0.076 0.070 •7.3 •0.153 0.068 •14.2 •0.120 0.056 •11.3 Copenhagen county •0.092 0.092 •8.8 •0.320 0.087 •27.4 •0.175 0.073 •16.1 Frederiksborg county 0.065 0.140 6.8 •0.086 0.144 •8.2 •0.065 0.175 •6.3 Roskilde county •0.039 0.139 •3.8 •0.152 0.126 •14.1 0.047 0.111 4.8 Western Zelland county •0.132 0.115 •12.4 •0.009 0.126 •0.9 0.019 0.144 2.0 Storstrøms county 0.112 0.131 11.8 •0.085 0.164 •8.1 0.073 0.136 7.6 Funen county •0.045 0.071 •4.4 •0.250 0.071 •22.1 •0.127 0.063 •11.9 Southern Jutland county •0.047 0.130 •4.6 •0.267 0.145 •23.4 •0.044 0.109 •4.3 Ribe county •0.011 0.132 •1.1 •0.212 0.126 •19.1 •0.071 0.119 •6.9 Vejle county •0.237 0.120 •21.1 •0.350 0.130 •29.5 •0.066 0.101 •6.3 Ringkøbing county •0.054 0.167 •5.3 •0.637 0.158 •47.1 •0.282 0.113 •24.6 Århus county •0.060 0.067 •5.8 •0.096 0.068 •9.1 •0.168 0.062 •15.5 Viborg county •0.703 0.139 •50.5 •0.887 0.150 •58.8 •0.116 0.114 •11.0 Northern Jutland county 0.032 0.073 3.3 •0.134 0.077 •12.5 •0.057 0.070 •5.6

53 TABLE A.3.11. POST-PROGRAMME EFFECTS OF EDUCATION TO WOMEN (1999-2001)

1999 2000 2001 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •0.005 0.073 •0.5 •0.059 0.070 •5.7 0.052 0.068 5.3 Copenhagen county •0.235 0.097 •20.9 •0.008 0.087 •0.8 0.146 0.080 15.7 Frederiksborg county •0.094 0.113 •9.0 0.057 0.101 5.8 •0.117 0.106 •11.0 Roskilde county 0.127 0.125 13.5 •0.015 0.124 •1.5 •0.162 0.147 •14.9 Western Zelland county •0.025 0.105 •2.4 •0.013 0.088 •1.3 0.075 0.098 7.7 Storstrøms county •0.120 0.115 •11.3 •0.012 0.102 •1.2 •0.026 0.110 •2.5 Funen county •0.024 0.080 •2.4 •0.082 0.074 •7.9 0.120 0.070 12.7 Southern Jutland county •0.076 0.110 •7.4 0.105 0.095 11.1 0.030 0.090 3.0 Ribe county •0.236 0.171 •21.0 0.062 0.137 6.4 0.077 0.108 8.0 Vejle county •0.047 0.091 •4.6 •0.132 0.082 •12.4 •0.092 0.093 •8.8 Ringkøbing county 0.053 0.090 5.4 •0.085 0.094 •8.1 0.001 0.096 0.1 Århus county •0.108 0.071 •10.2 0.012 0.062 1.2 0.048 0.053 4.9 Viborg county •0.063 0.127 •6.1 •0.083 0.122 •8.0 0.043 0.126 4.3 Northern Jutland county 0.034 0.074 3.5 0.069 0.069 7.1 0.111 0.068 11.8

TABLE A.3.12. POST-PROGRAMME EFFECTS OF EDUCATION TO WOMEN (2002-2004)

2002 2003 2004 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •0.002 0.077 •0.2 •0.147 0.074 •13.7 •0.220 0.061 •19.7 Copenhagen county •0.077 0.084 •7.4 •0.166 0.080 •15.3 •0.177 0.077 •16.3 Frederiksborg county 0.049 0.115 5.1 •0.194 0.145 •17.6 0.040 0.115 4.1 Roskilde county •0.307 0.134 •26.4 •0.037 0.113 •3.6 •0.073 0.110 •7.0 Western Zelland county 0.158 0.084 17.1 •0.169 0.116 •15.5 •0.315 0.163 •27.1 Storstrøms county 0.112 0.097 11.8 •0.015 0.118 •1.5 •0.022 0.138 •2.2 Funen county 0.039 0.069 3.9 •0.197 0.066 •17.9 •0.231 0.065 •20.6 Southern Jutland county •0.007 0.098 •0.7 •0.159 0.126 •14.7 •0.111 0.101 •10.5 Ribe county •0.007 0.106 •0.7 •0.186 0.128 •17.0 •0.216 0.137 •19.5 Vejle county •0.237 0.099 •21.1 •0.350 0.106 •29.5 •0.066 0.097 •6.3 Ringkøbing county 0.004 0.112 0.4 •0.135 0.124 •12.7 •0.231 0.110 •20.7 Århus county 0.030 0.056 3.0 •0.105 0.065 •10.0 •0.096 0.056 •9.2 Viborg county 0.068 0.105 7.1 •0.026 0.117 •2.5 •0.143 0.105 •13.3 Northern Jutland county 0.088 0.059 9.2 0.038 0.067 3.8 •0.002 0.076 •0.2

54 TABLE A.3.13. POST-PROGRAMME EFFECTS OF OTHER ALMPSs TO MEN (1999-2001)

1999 2000 2001 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg 0.032 0.138 3.2 •0.240 0.157 •21.3 •0.169 0.104 •15.6 Copenhagen county 0.317 0.170 37.3 •0.100 0.199 •9.5 •0.223 0.139 •20.0 Frederiksborg county •0.101 0.361 •9.6 0.255 0.240 29.1 •0.375 0.248 •31.3 Roskilde county 0.114 0.424 12.0 •0.783 0.458 •54.3 •0.154 0.265 •14.3 Western Zelland county •0.127 0.238 •11.9 0.056 0.236 5.7 •0.311 0.209 •26.7 Storstrøms county •0.306 0.261 •26.4 0.000 0.161 0.0 •0.152 0.188 •14.1 Funen county 0.247 0.599 28.0 •0.293 0.266 •25.4 0.021 0.140 2.1 Southern Jutland county 0.077 0.367 8.0 •0.320 0.361 •27.3 •0.217 0.285 •19.5 Ribe county •0.594 0.294 •44.8 •0.243 0.462 •21.5 0.677 0.344 96.8 Vejle county •0.103 0.277 •9.8 0.139 0.204 14.9 0.125 0.178 13.3 Ringkøbing county •0.418 0.323 •34.1 •0.203 0.368 •18.3 0.002 0.219 0.2 Århus county •0.203 0.190 •18.3 •0.099 0.121 •9.4 •0.130 0.124 •12.2 Viborg county •0.685 0.257 •49.6 •0.179 0.221 •16.4 •0.139 0.213 •13.0 Northern Jutland county •0.481 0.263 •38.2 •0.069 0.236 •6.6 0.082 0.246 8.5

TABLE A.3.14. POST-PROGRAMME EFFECTS OF OTHER ALMPSs TO MEN (2002-2004)

2002 2003 2004 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg 0.015 0.066 1.5 •0.338 0.052 •28.7 •0.142 0.076 •13.2 Copenhagen county •0.085 0.097 •8.2 •0.290 0.065 •25.2 •0.219 0.096 •19.7 Frederiksborg county •0.049 0.173 •4.8 •0.033 0.227 •3.2 •0.157 0.150 •14.5 Roskilde county •0.216 0.207 •19.4 •0.269 0.349 •23.6 •0.740 0.295 •52.3 Western Zelland county 0.085 0.184 8.8 •0.285 0.177 •24.8 •0.144 0.114 •13.4 Storstrøms county •0.198 0.140 •18.0 •0.401 0.096 •33.0 •0.394 0.096 •32.6 Funen county •0.014 0.106 •1.4 •0.172 0.227 •15.8 •0.360 0.161 •30.2 Southern Jutland county •0.327 0.275 •27.9 •0.110 0.299 •10.4 •0.156 0.194 •14.4 Ribe county •0.188 0.342 •17.1 •0.398 0.563 •32.8 •0.403 0.565 •33.1 Vejle county •0.110 0.144 •10.4 •0.389 0.104 •32.2 •0.195 0.094 •17.7 Ringkøbing county 0.135 0.126 14.5 •0.197 0.147 •17.9 •0.530 0.203 •41.1 Århus county •0.130 0.094 •12.2 •0.360 0.118 •30.2 •0.263 0.086 •23.1 Viborg county •0.004 0.166 •0.4 0.012 0.301 1.2 •0.229 0.259 •20.5 Northern Jutland county 0.040 0.163 4.1 •0.425 0.180 •34.6 •0.407 0.100 •33.4

55 TABLE A.3.15. POST-PROGRAMME EFFECTS OF OTHER ALMPSs TO WOMEN (1999-2001)

1999 2000 2001 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •0.021 0.150 •2.1 •0.174 0.152 •16.0 0.114 0.118 12.1 Copenhagen county •0.149 0.179 •13.8 0.136 0.215 14.5 0.003 0.130 0.3 Frederiksborg county 0.169 0.268 18.4 •0.328 0.268 •28.0 0.358 0.164 43.0 Roskilde county •0.267 0.223 •23.4 0.183 0.421 20.0 0.119 0.201 12.6 Western Zelland county •0.632 0.341 •46.8 0.013 0.230 1.3 •0.037 0.168 •3.6 Storstrøms county •0.793 0.278 •54.7 •0.301 0.159 •26.0 •0.026 0.180 •2.5 Funen county 0.098 0.414 10.3 •0.096 0.198 •9.1 •0.057 0.124 •5.6 Southern Jutland county •0.735 0.486 •52.1 •0.128 0.284 •12.0 0.144 0.194 15.5 Ribe county 0.106 0.221 11.1 •0.258 0.249 •22.7 •0.042 0.323 •4.1 Vejle county •0.103 0.179 •9.8 0.139 0.147 14.9 0.125 0.116 13.3 Ringkøbing county •0.288 0.175 •25.0 •0.054 0.157 •5.2 0.199 0.118 22.1 Århus county •0.040 0.133 •3.9 •0.048 0.117 •4.7 0.033 0.098 3.3 Viborg county •0.159 0.168 •14.7 •0.240 0.169 •21.3 •0.036 0.146 •3.6 Northern Jutland county •0.034 0.159 •3.4 •0.308 0.221 •26.5 •0.250 0.216 •22.1

TABLE A.3.16. POST-PROGRAMME EFFECTS OF OTHER ALMPSs TO WOMEN (2002-2004)

2002 2003 2004 Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Copenhagen Frederiksberg •0.113 0.077 •10.7 •0.368 0.054 •30.8 •0.229 0.077 •20.5 Copenhagen county •0.073 0.102 •7.1 •0.073 0.072 •7.1 •0.308 0.095 •26.5 Frederiksborg county 0.160 0.132 17.4 •0.183 0.212 •16.7 •0.011 0.337 •1.1 Roskilde county 0.327 0.167 38.7 0.199 0.247 22.1 0.243 0.260 27.5 Western Zelland county •0.034 0.309 •3.4 •0.078 0.194 •7.5 0.166 0.090 18.1 Storstrøms county 0.004 0.137 0.4 •0.365 0.102 •30.6 •0.185 0.091 •16.9 Funen county 0.055 0.096 5.7 •0.282 0.217 •24.6 •0.146 0.137 •13.6 Southern Jutland county 0.377 0.139 45.7 •0.481 0.242 •38.2 •0.041 0.153 •4.0 Ribe county 0.214 0.265 23.8 •0.028 0.267 •2.8 0.049 0.193 5.0 Vejle county •0.110 0.097 •10.4 •0.389 0.087 •32.2 •0.195 0.076 •17.7 Ringkøbing county •0.055 0.108 •5.3 •0.055 0.126 •5.3 •0.374 0.152 •31.2 Århus county 0.090 0.079 9.4 •0.105 0.121 •10.0 •0.194 0.084 •17.7 Viborg county •0.026 0.154 •2.6 •0.337 0.252 •28.6 •0.295 0.193 •25.5 Northern Jutland county 0.114 0.161 12.1 •0.043 0.149 •4.2 •0.276 0.102 •24.1

56 TABLE A.4.1. UNEMPLOYMENT DURATIONS (IN WEEKS) OF MEN ACTIVATED IN

PRIVATE JOB TRAINING PROGRAMMES (1999-2004)

1999 2000 2001 2002 2003 2004 Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Copenhagen Frederiksberg 22.84 1.02 23.29 1.04 22.13 0.99 21.81 0.98 21.61 0.97 23.19 1.04 Copenhagen county 20.25 0.91 19.68 0.88 21.05 0.94 19.04 0.85 19.32 0.86 20.29 0.91 Frederiksborg county • • 16.28 0.73 15.04 0.67 14.89 0.67 16.19 0.72 15.50 0.69 Roskilde county 16.82 0.75 19.53 0.87 17.09 0.76 16.86 0.75 17.33 0.78 18.96 0.85 Western Zelland county 14.21 0.64 15.27 0.68 14.12 0.63 15.47 0.69 14.27 0.64 14.89 0.67 Storstrøms county 14.48 0.65 13.61 0.61 13.57 0.61 13.19 0.59 14.31 0.64 13.87 0.62 Funen county 10.26 0.46 10.47 0.47 10.02 0.45 10.06 0.45 10.01 0.45 10.13 0.45 Southern Jutland county 14.11 0.63 13.91 0.62 13.59 0.61 12.03 0.54 13.18 0.59 13.45 0.60 Ribe county 8.93 0.40 9.87 0.44 9.50 0.43 9.99 0.45 9.81 0.44 9.92 0.44 Vejle county 12.58 0.56 12.45 0.56 12.25 0.55 12.57 0.56 12.12 0.54 12.07 0.54 Ringkøbing county 8.69 0.39 8.85 0.40 8.39 0.38 8.94 0.40 8.55 0.38 8.87 0.40 Århus county 13.76 0.62 13.77 0.62 13.59 0.61 13.23 0.59 13.70 0.61 13.49 0.60 Viborg county 8.67 0.39 9.10 0.41 8.75 0.39 8.48 0.38 8.52 0.38 8.51 0.38 Northern Jutland county 14.27 0.64 13.86 0.62 13.89 0.62 13.83 0.62 14.02 0.63 14.19 0.63

TABLE A.4.2. UNEMPLOYMENT DURATIONS (IN WEEKS) OF WOMEN ACTIVATED IN

PRIVATE JOB TRAINING PROGRAMMES (1999-2004)

1999 2000 2001 2002 2003 2004 Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Copenhagen Frederiksberg 33.69 1.51 33.66 1.51 34.25 1.53 34.62 1.55 31.21 1.40 32.17 1.44 Copenhagen county 33.56 1.50 31.55 1.41 32.69 1.46 27.77 1.24 30.93 1.38 31.40 1.40 Frederiksborg county 25.19 1.13 26.58 1.19 25.08 1.12 25.98 1.16 26.25 1.17 25.45 1.14 Roskilde county 20.20 0.90 19.25 0.86 20.56 0.92 19.17 0.86 21.57 0.96 23.74 1.06 Western Zelland county 23.46 1.05 26.82 1.20 26.87 1.20 26.44 1.18 26.74 1.20 27.95 1.25 Storstrøms county 22.43 1.00 20.11 0.90 21.78 0.97 22.29 1.00 22.72 1.02 21.86 0.98 Funen county 19.60 0.88 19.60 0.88 19.84 0.89 18.67 0.83 19.24 0.86 18.98 0.85 Southern Jutland county 23.66 1.06 23.76 1.06 23.99 1.07 22.65 1.01 25.61 1.15 23.12 1.03 Ribe county 25.52 1.14 27.05 1.21 26.74 1.20 26.48 1.18 25.65 1.15 26.13 1.17 Vejle county 22.23 0.99 20.49 0.92 20.32 0.91 19.42 0.87 19.47 0.87 21.29 0.95 Ringkøbing county 22.46 1.00 22.13 0.99 20.28 0.91 20.47 0.92 21.95 0.98 21.39 0.96 Århus county 24.22 1.08 22.34 1.00 23.04 1.03 21.97 0.98 22.33 1.00 22.59 1.01 Viborg county 19.91 0.89 17.11 0.77 17.90 0.80 16.49 0.74 17.33 0.78 18.06 0.81 Northern Jutland county 26.28 1.18 26.29 1.18 24.47 1.09 23.43 1.05 25.48 1.14 25.96 1.16

57 TABLE A.4.3. UNEMPLOYMENT DURATIONS (IN WEEKS) OF MEN ACTIVATED IN PUBLIC

JOB TRAINING PROGRAMMES (1999-2004)

1999 2000 2001 2002 2003 2004 Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Copenhagen Frederiksberg 24.52 1.10 24.49 1.10 24.29 1.09 24.76 1.11 24.34 1.09 24.60 1.10 Copenhagen county 22.25 1.00 22.51 1.01 22.85 1.02 21.27 0.95 21.87 0.98 22.01 0.98 Frederiksborg county 17.46 0.78 17.14 0.77 17.29 0.77 18.15 0.81 16.17 0.72 17.05 0.76 Roskilde county 20.02 0.90 19.13 0.86 19.75 0.88 21.40 0.96 20.66 0.92 18.85 0.84 Western Zelland county 16.80 0.75 16.84 0.75 16.22 0.73 15.86 0.71 16.93 0.76 16.36 0.73 Storstrøms county 15.16 0.68 14.73 0.66 14.66 0.66 14.87 0.66 15.57 0.70 15.14 0.68 Funen county 10.88 0.49 10.87 0.49 11.22 0.50 11.05 0.49 11.14 0.50 10.83 0.48 Southern Jutland county 15.38 0.69 14.02 0.63 14.03 0.63 14.82 0.66 14.53 0.65 14.35 0.64 Ribe county 10.97 0.49 10.25 0.46 10.55 0.47 10.46 0.47 10.50 0.47 10.10 0.45 Vejle county 13.02 0.58 13.06 0.58 14.48 0.65 13.72 0.61 13.00 0.58 13.63 0.61 Ringkøbing county 9.28 0.41 9.08 0.41 9.58 0.43 9.74 0.44 9.56 0.43 9.62 0.43 Århus county 15.08 0.67 14.81 0.66 14.72 0.66 14.42 0.65 14.75 0.66 14.28 0.64 Viborg county • • 8.93 0.40 9.33 0.42 8.95 0.40 9.08 0.41 9.06 0.41 Northern Jutland county 15.25 0.68 15.57 0.70 14.79 0.66 15.16 0.68 15.21 0.68 15.41 0.69

TABLE A.4.4. UNEMPLOYMENT DURATIONS (IN WEEKS) OF WOMEN ACTIVATED IN

PUBLIC JOB TRAINING PROGRAMMES (1999-2004)

1999 2000 2001 2002 2003 2004 Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Copenhagen Frederiksberg 39.49 1.77 36.80 1.65 36.88 1.65 37.35 1.67 35.68 1.60 36.91 1.65 Copenhagen county 35.78 1.60 34.45 1.54 34.49 1.54 34.04 1.52 35.28 1.58 33.39 1.49 Frederiksborg county 28.44 1.27 27.92 1.25 29.07 1.30 26.89 1.20 26.11 1.17 27.47 1.23 Roskilde county 24.81 1.11 24.09 1.08 24.48 1.09 24.18 1.08 25.39 1.14 24.64 1.10 Western Zelland county 27.17 1.22 28.02 1.25 28.18 1.26 28.16 1.26 28.18 1.26 28.38 1.27 Storstrøms county 24.23 1.08 23.63 1.06 23.56 1.05 24.78 1.11 23.75 1.06 24.77 1.11 Funen county 21.79 0.97 21.74 0.97 21.22 0.95 22.42 1.00 22.24 0.99 21.94 0.98 Southern Jutland county 27.23 1.22 26.80 1.20 27.99 1.25 26.70 1.19 26.76 1.20 26.53 1.19 Ribe county 28.41 1.27 28.40 1.27 27.21 1.22 27.71 1.24 27.14 1.21 28.62 1.28 Vejle county 23.75 1.06 22.09 0.99 23.23 1.04 22.45 1.00 23.46 1.05 23.60 1.06 Ringkøbing county 24.94 1.12 25.09 1.12 23.32 1.04 24.24 1.08 24.22 1.08 23.45 1.05 Århus county 26.51 1.19 24.98 1.12 24.50 1.10 24.69 1.10 25.69 1.15 24.69 1.10 Viborg county 19.85 0.89 19.97 0.89 18.86 0.84 18.96 0.85 18.55 0.83 18.56 0.83 Northern Jutland county 29.63 1.33 29.43 1.32 28.70 1.28 28.72 1.28 28.49 1.27 27.70 1.24

58 TABLE A.4.5. UNEMPLOYMENT DURATIONS (IN WEEKS) OF MEN ACTIVATED IN

EDUCATION PROGRAMMES (1999-2004)

1999 2000 2001 2002 2003 2004 Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Copenhagen Frederiksberg 23.99 1.07 23.94 1.07 23.43 1.05 23.96 1.07 24.19 1.08 24.03 1.07 Copenhagen county 21.18 0.95 21.14 0.95 21.15 0.95 21.29 0.95 21.65 0.97 21.46 0.96 Frederiksborg county 16.62 0.74 16.60 0.74 16.28 0.73 16.73 0.75 16.18 0.72 16.33 0.73 Roskilde county 18.92 0.85 18.82 0.84 19.65 0.88 19.16 0.86 19.16 0.86 19.13 0.86 Western Zelland county 15.68 0.70 15.57 0.70 15.25 0.68 15.68 0.70 15.65 0.70 15.73 0.70 Storstrøms county 14.73 0.66 14.22 0.64 14.27 0.64 14.28 0.64 14.43 0.65 14.24 0.64 Funen county 10.77 0.48 10.51 0.47 10.56 0.47 10.64 0.48 10.68 0.48 10.63 0.48 Southern Jutland county 14.23 0.64 14.10 0.63 14.01 0.63 14.22 0.64 14.54 0.65 14.09 0.63 Ribe county 10.36 0.46 10.21 0.46 10.26 0.46 10.14 0.45 10.40 0.47 10.19 0.46 Vejle county 13.55 0.61 13.19 0.59 13.07 0.58 13.23 0.59 13.72 0.61 13.38 0.60 Ringkøbing county 9.39 0.42 9.28 0.41 9.27 0.41 9.29 0.42 9.61 0.43 9.40 0.42 Århus county 14.66 0.66 14.48 0.65 14.48 0.65 14.61 0.65 14.54 0.65 14.69 0.66 Viborg county 9.02 0.40 8.79 0.39 9.02 0.40 8.82 0.39 9.01 0.40 8.90 0.40 Northern Jutland county 15.04 0.67 14.97 0.67 14.58 0.65 15.09 0.67 15.15 0.68 14.98 0.67

TABLE A.4.6. UNEMPLOYMENT DURATIONS (IN WEEKS) OF WOMEN ACTIVATED IN

EDUCATION PROGRAMMES (1999-2004)

1999 2000 2001 2002 2003 2004 Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Copenhagen Frederiksberg 37.08 1.66 37.16 1.66 36.67 1.64 37.04 1.66 37.53 1.68 37.84 1.69 Copenhagen county 34.86 1.56 34.27 1.53 33.32 1.49 34.17 1.53 34.55 1.54 34.86 1.56 Frederiksborg county 27.69 1.24 27.14 1.21 27.53 1.23 27.13 1.21 27.56 1.23 26.94 1.20 Roskilde county 24.12 1.08 24.21 1.08 24.35 1.09 24.84 1.11 24.11 1.08 24.25 1.08 Western Zelland county 27.76 1.24 27.55 1.23 26.88 1.20 26.74 1.20 27.83 1.24 28.28 1.26 Storstrøms county 24.40 1.09 23.88 1.07 23.65 1.06 23.49 1.05 23.71 1.06 24.18 1.08 Funen county 21.51 0.96 21.40 0.96 20.86 0.93 21.07 0.94 21.41 0.96 21.38 0.96 Southern Jutland county 26.51 1.19 25.94 1.16 26.02 1.16 26.18 1.17 26.69 1.19 26.48 1.18 Ribe county 28.19 1.26 26.92 1.20 26.94 1.20 27.16 1.21 27.56 1.23 28.06 1.25 Vejle county 23.24 1.04 23.23 1.04 23.06 1.03 23.43 1.05 23.92 1.07 22.95 1.03 Ringkøbing county 24.12 1.08 24.24 1.08 24.06 1.08 23.81 1.06 24.33 1.09 24.84 1.11 Århus county 25.67 1.15 25.34 1.13 25.30 1.13 25.18 1.13 25.43 1.14 25.56 1.14 Viborg county 19.29 0.86 19.01 0.85 18.54 0.83 18.62 0.83 18.90 0.85 19.23 0.86 Northern Jutland county 28.62 1.28 28.02 1.25 28.14 1.26 28.31 1.27 28.39 1.27 28.68 1.28

59 TABLE A.4.7. UNEMPLOYMENT DURATIONS (IN WEEKS) OF MEN ACTIVATED IN OTHER

ALMPs (1999-2004)

1999 2000 2001 2002 2003 2004 Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Copenhagen Frederiksberg 23.01 1.03 23.52 1.05 23.17 1.04 22.58 1.01 23.36 1.04 23.15 1.04 Copenhagen county 19.69 0.88 20.66 0.92 21.03 0.94 20.51 0.92 20.96 0.94 20.64 0.92 Frederiksborg county 16.19 0.72 15.92 0.71 16.85 0.75 16.38 0.73 16.07 0.72 16.48 0.74 Roskilde county 18.43 0.82 19.55 0.87 18.64 0.83 18.82 0.84 19.24 0.86 19.92 0.89 Western Zelland county 15.48 0.69 14.62 0.65 15.76 0.70 15.19 0.68 15.48 0.69 15.34 0.70 Storstrøms county 14.50 0.65 13.89 0.62 14.01 0.63 14.15 0.63 14.36 0.64 14.26 0.64 Funen county 9.89 0.44 10.29 0.46 10.28 0.46 10.28 0.46 10.50 0.47 10.61 0.47 Southern Jutland county 13.81 0.62 13.65 0.61 14.07 0.63 14.15 0.63 13.34 0.60 13.80 0.62 Ribe county 10.17 0.45 10.12 0.45 9.22 0.41 9.84 0.44 10.53 0.47 10.37 0.46 Vejle county 12.86 0.58 12.50 0.56 12.52 0.56 12.92 0.58 13.11 0.59 12.92 0.58 Ringkøbing county 9.34 0.42 9.12 0.41 8.95 0.40 8.91 0.40 9.13 0.41 9.10 0.41 Århus county 14.16 0.63 13.86 0.62 13.99 0.63 13.93 0.62 14.38 0.64 14.04 0.63 Viborg county 8.88 0.40 8.71 0.39 8.56 0.38 8.60 0.38 8.49 0.38 8.83 0.39 Northern Jutland county 14.73 0.66 14.35 0.64 14.49 0.65 14.56 0.65 15.08 0.67 14.84 0.66

TABLE A.4.8. UNEMPLOYMENT DURATIONS (IN WEEKS) OF WOMEN ACTIVATED IN

OTHER ALMPs (1999-2004)

1999 2000 2001 2002 2003 2004 Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Duration Std. Err. Copenhagen Frederiksberg 35.72 1.60 36.31 1.62 35.56 1.59 35.42 1.58 36.99 1.65 36.40 1.63 Copenhagen county 33.47 1.50 31.93 1.43 32.12 1.44 32.52 1.45 34.08 1.52 33.62 1.50 Frederiksborg county 25.29 1.13 27.08 1.21 24.51 1.10 25.38 1.14 26.99 1.21 26.39 1.18 Roskilde county 23.67 1.06 21.57 0.96 21.58 0.97 21.56 0.96 22.93 1.03 22.15 0.99 Western Zelland county 27.18 1.22 26.69 1.19 25.71 1.15 26.03 1.16 26.18 1.17 25.52 1.14 Storstrøms county 24.41 1.09 23.38 1.05 22.43 1.00 22.98 1.03 23.39 1.05 23.33 1.04 Funen county 20.37 0.91 20.15 0.90 20.72 0.93 20.06 0.90 20.12 0.90 20.20 0.90 Southern Jutland county 26.41 1.18 25.50 1.14 24.24 1.08 23.91 1.07 25.05 1.12 24.22 1.08 Ribe county 25.77 1.15 27.08 1.21 26.48 1.18 24.73 1.11 25.62 1.15 26.25 1.17 Vejle county 22.41 1.00 21.28 0.95 21.26 0.95 21.71 0.97 22.87 1.02 22.34 1.00 Ringkøbing county 23.93 1.07 23.46 1.05 22.47 1.00 23.09 1.03 23.47 1.05 23.80 1.06 Århus county 24.30 1.09 23.92 1.07 23.94 1.07 23.78 1.06 25.02 1.12 24.41 1.09 Viborg county 18.66 0.83 18.65 0.83 17.86 0.80 18.27 0.82 18.51 0.83 18.47 0.83 Northern Jutland county 27.51 1.23 28.49 1.27 28.41 1.27 26.55 1.19 27.64 1.24 27.59 1.23

60 TABLE A.4.9. UNEMPLOYMENT DURATIONS (IN WEEKS) WITHOUT ACTIVATION

Men Women Duration Std. Err. Duration Std. Err. Copenhagen Frederiksberg 22.52 1.01 34.60 1.55 Copenhagen county 20.15 0.90 31.91 1.43 Frederiksborg county 15.91 0.71 25.48 1.14 Roskilde county 18.32 0.82 22.33 1.00 Western Zelland county 15.05 0.67 25.67 1.15 Storstrøms county 13.79 0.62 22.35 1.00 Funen county 10.19 0.46 19.75 0.88 Southern Jutland county 13.59 0.61 24.42 1.09 Ribe county 9.85 0.44 25.42 1.14 Vejle county 12.52 0.56 21.36 0.96 Ringkøbing county 8.88 0.40 22.38 1.00 Århus county 13.79 0.62 23.62 1.06 Viborg county 8.50 0.38 17.60 0.79 Northern Jutland county 14.31 0.64 26.65 1.19

61 TABLE A.5.1. NET EFFECTS OF PRIVATE JOB TRAINING TO MEN (1999-2004)

1999 2000 2001 2002 2003 2004 Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Copenhagen Frederiksberg 0.317 0.153 0.765 0.272 •0.394 0.148 •0.709 0.343 •0.918 0.144 0.668 0.144 Copenhagen county 0.101 0.068 •0.466 0.335 0.907 0.600 •1.104 0.501 •0.831 0.377 0.147 0.029 Frederiksborg county • • 0.370 0.138 •0.871 0.413 •1.014 0.490 0.284 0.451 •0.403 0.119 Roskilde county •1.502 0.791 1.208 0.869 •1.233 0.469 •1.464 0.662 •0.989 0.612 0.636 0.231 Western Zelland county •0.843 1.144 0.224 0.209 •0.934 0.319 0.416 0.456 •0.786 0.243 •0.163 0.060 Storstrøms county 0.698 0.651 •0.173 0.041 •0.216 0.048 •0.594 0.147 0.521 0.391 0.089 0.028 Funen county 0.071 0.020 0.277 0.080 •0.174 0.044 •0.135 0.036 •0.177 0.038 •0.061 0.009 Southern Jutland county 0.523 0.249 0.326 0.228 0.002 0.001 •1.559 0.432 •0.406 0.241 •0.143 0.106 Ribe county •0.923 0.848 0.018 0.006 •0.351 0.110 0.139 0.091 •0.047 0.015 0.069 0.014 Vejle county 0.064 0.041 •0.073 0.033 •0.268 0.095 0.045 0.010 •0.401 0.098 •0.453 0.577 Ringkøbing county •0.185 0.063 •0.032 0.014 •0.492 0.162 0.057 0.060 •0.331 0.138 •0.010 0.002 Århus county •0.033 0.004 •0.015 0.004 •0.201 0.032 •0.554 0.127 •0.087 0.016 •0.297 0.035 Viborg county 0.177 0.103 0.608 0.331 0.250 0.086 •0.012 0.005 0.020 0.010 0.012 0.003 Northern Jutland county •0.036 0.013 •0.441 0.077 •0.417 0.063 •0.473 0.062 •0.288 0.081 •0.112 0.019

TABLE A.5.2. NET EFFECTS OF PRIVATE JOB TRAINING TO WOMEN (1999-2004)

1999 2000 2001 2002 2003 2004 Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Copenhagen Frederiksberg •0.911 0.316 •0.941 1.216 •0.352 0.164 0.014 0.004 •3.395 0.966 •2.430 0.368 Copenhagen county 1.646 3.488 •0.357 0.620 0.776 1.255 •4.138 1.153 •0.982 0.389 •0.513 0.227 Frederiksborg county •0.293 0.294 1.095 0.605 •0.397 0.107 0.497 0.453 0.769 0.965 •0.029 0.011 Roskilde county •2.133 4.073 •3.086 4.392 •1.769 2.684 •3.158 1.121 •0.762 6.018 1.412 0.492 Western Zelland county •2.202 2.845 1.154 0.834 1.208 1.472 0.772 0.462 1.079 1.058 2.280 0.930 Storstrøms county 0.080 0.030 •2.235 0.770 •0.564 0.191 •0.062 0.026 0.368 0.143 •0.486 0.084 Funen county •0.149 0.043 •0.152 0.036 0.091 0.024 •1.083 0.344 •0.512 0.140 •0.772 0.162 Southern Jutland county •0.767 2.016 •0.667 0.314 •0.431 0.250 •1.775 0.601 1.187 0.745 •1.303 0.422 Ribe county 0.099 0.038 1.627 1.205 1.326 1.065 1.060 1.422 0.237 0.317 0.710 0.207 Vejle county 0.875 0.404 •0.867 0.564 •1.038 0.285 •1.942 0.372 •1.888 0.440 •0.067 0.102 Ringkøbing county 0.078 0.031 •0.255 0.571 •2.106 0.573 •1.916 0.721 •0.431 0.393 •0.994 0.410 Århus county 0.604 0.121 •1.270 0.209 •0.570 0.111 •1.644 0.419 •1.289 0.302 •1.030 0.170 Viborg county 2.312 1.424 •0.491 0.329 0.306 0.610 •1.107 0.463 •0.266 0.182 0.465 0.216 Northern Jutland county •0.371 0.104 •0.358 0.073 •2.175 0.516 •3.217 0.543 •1.165 0.230 •0.683 0.307

62 TABLE A.5.3. NET EFFECTS OF PUBLIC JOB TRAINING TO MEN (1999-2004)

1999 2000 2001 2002 2003 2004 Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Copenhagen Frederiksberg 1.997 0.558 1.965 0.388 1.768 0.340 2.235 0.335 1.819 0.262 2.078 0.328 Copenhagen county 2.105 0.450 2.364 0.562 2.706 0.468 1.120 0.231 1.724 0.430 1.863 0.346 Frederiksborg county 1.555 0.441 1.231 0.481 1.379 0.407 2.245 0.507 0.258 0.100 1.145 0.465 Roskilde county 1.699 2.283 0.811 0.242 1.431 0.570 3.081 0.916 2.335 0.943 0.525 0.116 Western Zelland county 1.747 0.512 1.792 0.507 1.173 0.497 0.813 0.274 1.881 0.528 1.305 0.380 Storstrøms county 1.372 0.376 0.949 0.371 0.878 0.188 1.082 0.257 1.780 0.305 1.357 0.207 Funen county 0.690 0.195 0.681 0.144 1.029 0.168 0.862 0.172 0.949 0.141 0.643 0.123 Southern Jutland county 1.794 0.582 0.427 0.087 0.443 0.175 1.234 0.419 0.936 0.387 0.764 0.273 Ribe county 1.113 0.372 0.396 0.133 0.692 0.370 0.604 0.291 0.644 0.287 0.248 0.146 Vejle county 0.501 0.401 0.540 0.282 1.954 0.998 1.203 1.221 0.477 0.209 1.111 0.420 Ringkøbing county 0.398 0.245 0.198 0.103 0.699 0.149 0.863 0.453 0.684 0.271 0.740 0.241 Århus county 1.287 0.395 1.019 0.287 0.930 0.167 0.637 0.186 0.961 0.213 0.493 0.155 Viborg county • • 0.434 0.152 0.837 0.221 0.457 0.282 0.583 0.332 0.561 0.223 Northern Jutland county 0.948 0.369 1.261 0.350 0.485 0.082 0.859 0.237 0.904 0.232 1.101 0.504

TABLE A.5.4. NET EFFECTS OF PUBLIC JOB TRAINING TO WOMEN (1999-2004)

1999 2000 2001 2002 2003 2004 Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Copenhagen Frederiksberg 4.886 1.127 2.196 0.377 2.273 0.515 2.745 0.349 1.076 0.161 2.311 0.410 Copenhagen county 3.872 0.939 2.540 0.175 2.583 0.643 2.129 0.462 3.374 0.641 1.476 0.255 Frederiksborg county 2.959 0.563 2.434 0.874 3.590 0.759 1.404 0.397 0.627 0.524 1.992 0.801 Roskilde county 2.476 0.969 1.761 0.746 2.147 0.702 1.852 0.717 3.061 0.782 2.312 0.617 Western Zelland county 1.506 0.238 2.356 0.629 2.511 0.476 2.493 0.625 2.512 0.756 2.718 0.605 Storstrøms county 1.884 0.427 1.278 0.519 1.211 0.248 2.434 0.469 1.402 0.393 2.419 0.406 Funen county 2.041 0.390 1.983 0.257 1.468 0.310 2.662 0.334 2.486 0.320 2.189 0.300 Southern Jutland county 2.810 0.872 2.376 0.740 3.566 0.618 2.273 0.548 2.335 0.264 2.109 0.519 Ribe county 2.992 0.613 2.985 0.721 1.792 0.401 2.293 0.660 1.721 0.746 3.202 0.654 Vejle county 2.389 0.820 0.733 0.141 1.870 0.436 1.091 0.598 2.105 0.521 2.244 0.464 Ringkøbing county 2.557 0.550 2.707 0.575 0.937 0.202 1.861 0.553 1.840 0.304 1.070 0.286 Århus county 2.891 0.609 1.360 0.241 0.880 0.146 1.080 1.036 2.075 0.340 1.077 0.222 Viborg county 2.248 0.544 2.371 0.583 1.259 0.348 1.360 0.435 0.955 0.512 0.964 0.194 Northern Jutland county 2.983 0.538 2.788 0.478 2.051 0.374 2.076 0.507 1.842 0.462 1.052 0.255

63 TABLE A.5.5. NET EFFECTS OF EDUCATION TO MEN (1999-2004)

1999 2000 2001 2002 2003 2004 Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Copenhagen Frederiksberg 1.466 0.172 1.416 0.153 0.903 0.070 1.436 0.130 1.667 0.146 1.511 0.133 Copenhagen county 1.033 0.166 0.991 0.154 1.006 0.152 1.140 0.163 1.505 0.163 1.310 0.156 Frederiksborg county 0.711 0.112 0.688 0.146 0.374 0.058 0.824 0.129 0.277 0.195 0.425 0.179 Roskilde county 0.599 0.153 0.503 0.127 1.330 0.295 0.841 0.268 0.839 0.215 0.812 0.189 Western Zelland county 0.628 0.207 0.523 0.113 0.203 0.047 0.633 0.127 0.604 0.152 0.681 0.182 Storstrøms county 0.949 0.167 0.433 0.081 0.482 0.079 0.495 0.091 0.650 0.207 0.452 0.105 Funen county 0.581 0.072 0.315 0.048 0.371 0.039 0.451 0.042 0.489 0.045 0.436 0.045 Southern Jutland county 0.639 0.212 0.508 0.066 0.424 0.077 0.632 0.145 0.953 0.166 0.496 0.151 Ribe county 0.508 0.171 0.355 0.094 0.408 0.060 0.282 0.089 0.548 0.101 0.336 0.083 Vejle county 1.028 0.182 0.670 0.089 0.550 0.076 0.704 0.087 1.198 0.129 0.863 0.167 Ringkøbing county 0.509 0.085 0.398 0.065 0.388 0.069 0.407 0.082 0.729 0.077 0.520 0.063 Århus county 0.868 0.091 0.687 0.058 0.694 0.052 0.820 0.081 0.748 0.075 0.902 0.073 Viborg county 0.526 0.114 0.296 0.073 0.523 0.035 0.327 0.035 0.518 0.054 0.402 0.080 Northern Jutland county 0.734 0.090 0.666 0.075 0.277 0.027 0.787 0.061 0.840 0.082 0.676 0.083

TABLE A.5.6. NET EFFECTS OF EDUCATION TO WOMEN (1999-2004)

1999 2000 2001 2002 2003 2004 Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Copenhagen Frederiksberg 2.482 0.261 2.556 0.233 2.067 0.181 2.433 0.214 2.923 0.224 3.238 0.220 Copenhagen county 2.946 0.290 2.360 0.245 1.407 0.126 2.264 0.216 2.635 0.258 2.954 0.257 Frederiksborg county 2.208 0.237 1.662 0.174 2.047 0.209 1.647 0.233 2.075 0.335 1.462 0.300 Roskilde county 1.784 0.233 1.874 0.311 2.014 0.390 2.505 0.327 1.774 0.315 1.918 0.330 Western Zelland county 2.091 0.301 1.879 0.266 1.218 0.170 1.073 0.114 2.169 0.298 2.610 0.387 Storstrøms county 2.048 0.254 1.531 0.199 1.299 0.220 1.140 0.135 1.360 0.294 1.833 0.254 Funen county 1.760 0.172 1.651 0.147 1.110 0.084 1.317 0.099 1.654 0.129 1.629 0.136 Southern Jutland county 2.088 0.246 1.515 0.168 1.600 0.175 1.751 0.215 2.267 0.284 2.053 0.271 Ribe county 2.774 0.399 1.505 0.269 1.519 0.194 1.739 0.226 2.146 0.314 2.639 0.359 Vejle county 1.880 0.200 1.874 0.199 1.701 0.187 2.076 0.206 2.566 0.245 1.591 0.247 Ringkøbing county 1.733 0.179 1.854 0.201 1.678 0.194 1.427 0.213 1.943 0.256 2.458 0.231 Århus county 2.058 0.166 1.729 0.132 1.685 0.101 1.569 0.122 1.814 0.149 1.949 0.148 Viborg county 1.694 0.227 1.414 0.233 0.938 0.169 1.018 0.135 1.299 0.218 1.632 0.205 Northern Jutland county 1.972 0.163 1.377 0.133 1.497 0.107 1.664 0.102 1.749 0.139 2.038 0.201

64 TABLE A.5.7. NET EFFECTS OF OTHER ALMPSs TO MEN (1999-2004)

1999 2000 2001 2002 2003 2004 Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Copenhagen Frederiksberg 0.488 0.157 0.997 0.284 0.651 0.177 0.060 0.034 0.841 0.097 0.630 0.158 Copenhagen county •0.459 0.131 0.513 0.321 0.885 0.191 0.362 0.143 0.814 0.115 0.494 0.119 Frederiksborg county 0.279 0.403 0.017 0.005 0.943 0.223 0.477 0.145 0.166 0.294 0.577 0.171 Roskilde county 0.111 0.090 1.225 0.616 0.318 0.365 0.496 0.285 0.914 0.513 1.600 0.394 Western Zelland county 0.427 0.664 •0.429 0.333 0.707 0.203 0.136 0.060 0.425 0.191 0.286 0.120 Storstrøms county 0.712 0.217 0.106 0.119 0.227 0.181 0.364 0.126 0.572 0.094 0.479 0.092 Funen county •0.301 0.301 0.102 0.058 0.091 0.041 0.087 0.052 0.311 0.138 0.417 0.089 Southern Jutland county 0.222 0.183 0.065 0.034 0.484 0.235 0.557 0.231 •0.245 0.136 0.210 0.219 Ribe county 0.316 0.089 0.262 0.222 •0.631 0.215 •0.017 0.011 0.673 0.891 0.516 0.201 Vejle county 0.342 0.090 •0.018 0.010 •0.003 0.001 0.402 0.325 0.593 0.086 0.401 0.075 Ringkøbing county 0.459 0.129 0.236 0.158 0.069 0.061 0.031 0.010 0.246 0.077 0.219 0.069 Århus county 0.369 0.172 0.067 0.051 0.204 0.079 0.143 0.087 0.589 0.125 0.256 0.068 Viborg county 0.386 0.107 0.214 0.110 0.069 0.097 0.107 0.082 •0.009 0.146 0.331 0.124 Northern Jutland county 0.427 0.184 0.043 0.097 0.185 0.090 0.249 0.107 0.771 0.178 0.539 0.100

TABLE A.5.8. NET EFFECTS OF OTHER ALMPSs TO WOMEN (1999-2004)

1999 2000 2001 2002 2003 2004 Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Net Eff. Std. Err. Copenhagen Frederiksberg 1.120 0.386 1.710 0.483 0.962 0.328 0.815 0.235 2.385 0.183 1.793 0.269 Copenhagen county 1.564 0.461 0.025 0.014 0.205 0.256 0.608 0.280 2.166 0.380 1.708 0.296 Frederiksborg county •0.193 0.145 1.603 0.721 •0.969 0.313 •0.101 0.034 1.506 0.522 0.913 0.262 Roskilde county 1.340 0.425 •0.766 0.988 •0.750 0.375 •0.776 0.305 0.600 0.318 •0.181 0.178 Western Zelland county 1.516 0.749 1.019 0.451 0.046 0.106 0.368 0.420 0.514 0.386 •0.146 0.036 Storstrøms county 2.060 0.467 1.031 0.321 0.081 0.095 0.627 0.194 1.044 0.214 0.985 0.182 Funen county 0.617 0.347 0.400 0.374 0.963 0.184 0.307 0.084 0.364 0.208 0.442 0.274 Southern Jutland county 1.984 0.952 1.073 0.474 •0.183 0.108 •0.514 0.089 0.622 0.198 •0.201 0.170 Ribe county 0.355 0.144 1.665 0.433 1.066 0.449 •0.689 0.720 0.201 0.706 0.832 0.371 Vejle county 1.056 0.231 •0.074 0.032 •0.095 0.033 0.348 0.206 1.513 0.180 0.987 0.158 Ringkøbing county 1.544 0.280 1.073 0.311 0.083 0.015 0.707 0.145 1.083 0.321 1.417 0.314 Århus county 0.689 0.251 0.302 0.145 0.321 0.121 0.164 0.041 1.406 0.764 0.793 0.130 Viborg county 1.061 0.233 1.049 0.311 0.258 0.218 0.677 0.214 0.915 0.418 0.875 0.312 Northern Jutland county 0.866 0.504 1.849 0.426 1.766 0.432 •0.093 0.057 0.997 0.305 0.947 0.259

65 TABLE A.6.1. EFFECTS (%) OF PRIVATE JOB TRAINING TO UNEMPLOYMENT DURATION

OF MEN (1999-2004)

Men 1999 2000 2001 2002 2003 2004 Copenhagen Frederiksberg 1.41 3.39 •1.75 •3.15 •4.07 2.97 Copenhagen county 0.50 •2.31 4.50 •5.48 •4.12 0.73 Frederiksborg county • 2.33 •5.47 •6.38 1.78 •2.53 Roskilde county •8.20 6.59 •6.73 •7.99 •5.40 3.47 Western Zelland county •5.60 1.49 •6.20 2.77 •5.22 •1.09 Storstrøms county 5.07 •1.26 •1.57 •4.31 3.78 0.65 Funen county 0.69 2.71 •1.71 •1.33 •1.74 •0.60 Southern Jutland county 3.85 2.40 0.02 •11.48 •2.99 •1.05 Ribe county •9.37 0.18 •3.56 1.41 •0.47 0.70 Vejle county 0.51 •0.58 •2.14 0.36 •3.20 •3.62 Ringkøbing county •2.08 •0.36 •5.54 0.64 •3.72 •0.11 Århus county •0.24 •0.11 •1.46 •4.02 •0.63 •2.15 Viborg county 2.08 7.16 2.94 •0.14 0.24 0.14 Northern Jutland county •0.25 •3.08 •2.91 •3.31 •2.01 •0.78

TABLE A.6.2. EFFECTS (%) OF PRIVATE JOB TRAINING TO UNEMPLOYMENT DURATION

OF WOMEN (1999-2004)

Women 1999 2000 2001 2002 2003 2004 Copenhagen Frederiksberg •2.63 •2.72 •1.02 0.04 •9.81 •7.02 Copenhagen county 5.16 •1.12 2.43 •12.97 •3.08 •1.61 Frederiksborg county •1.15 4.30 •1.56 1.95 3.02 •0.11 Roskilde county •9.55 •13.82 •7.92 •14.14 •3.41 6.32 Western Zelland county •8.58 4.50 4.71 3.01 4.20 8.88 Storstrøms county 0.36 •10.00 •2.52 •0.28 1.65 •2.17 Funen county •0.75 •0.77 0.46 •5.48 •2.59 •3.91 Southern Jutland county •3.14 •2.73 •1.76 •7.27 4.86 •5.34 Ribe county 0.39 6.40 5.22 4.17 0.93 2.79 Vejle county 4.10 •4.06 •4.86 •9.09 •8.84 •0.31 Ringkøbing county 0.35 •1.14 •9.41 •8.56 •1.93 •4.44 Århus county 2.56 •5.38 •2.41 •6.96 •5.46 •4.36 Viborg county 13.14 •2.79 1.74 •6.29 •1.51 2.64 Northern Jutland county •1.39 •1.34 •8.16 •12.07 •4.37 •2.56

66 TABLE A.6.3. EFFECTS (%) OF PUBLIC JOB TRAINING TO UNEMPLOYMENT DURATION

OF MEN (1999-2004)

Men 1999 2000 2001 2002 2003 2004 Copenhagen Frederiksberg 8.86 8.72 7.85 9.92 8.08 9.23 Copenhagen county 10.45 11.74 13.43 5.56 8.56 9.25 Frederiksborg county 9.77 7.74 8.67 14.12 1.62 7.20 Roskilde county 9.28 4.43 7.81 16.82 12.74 2.87 Western Zelland county 11.61 11.90 7.79 5.40 12.50 8.67 Storstrøms county 9.95 6.88 6.37 7.85 12.92 9.84 Funen county 6.77 6.68 10.10 8.46 9.31 6.31 Southern Jutland county 13.20 3.14 3.26 9.08 6.89 5.62 Ribe county 11.30 4.02 7.02 6.13 6.53 2.52 Vejle county 4.00 4.31 15.61 9.61 3.81 8.87 Ringkøbing county 4.49 2.23 7.88 9.72 7.71 8.33 Århus county 9.34 7.39 6.74 4.62 6.97 3.58 Viborg county • 5.11 9.85 5.38 6.87 6.60 Northern Jutland county 6.62 8.82 3.39 6.01 6.32 7.70

TABLE A.6.4. EFFECTS (%) OF PUBLIC JOB TRAINING TO UNEMPLOYMENT DURATION

OF WOMEN (1999-2004)

Women 1999 2000 2001 2002 2003 2004 Copenhagen Frederiksberg 14.12 6.35 6.57 7.93 3.11 6.68 Copenhagen county 12.13 7.96 8.10 6.67 10.57 4.63 Frederiksborg county 11.61 9.55 14.09 5.51 2.46 7.82 Roskilde county 11.09 7.88 9.61 8.30 13.71 10.35 Western Zelland county 5.87 9.18 9.78 9.71 9.79 10.59 Storstrøms county 8.43 5.72 5.42 10.89 6.27 10.82 Funen county 10.33 10.04 7.43 13.47 12.58 11.08 Southern Jutland county 11.51 9.73 14.60 9.31 9.56 8.64 Ribe county 11.77 11.74 7.05 9.02 6.77 12.60 Vejle county 11.18 3.43 8.76 5.11 9.86 10.51 Ringkøbing county 11.42 12.09 4.18 8.31 8.22 4.78 Århus county 12.24 5.76 3.73 4.57 8.79 4.56 Viborg county 12.77 13.47 7.16 7.73 5.43 5.48 Northern Jutland county 11.19 10.46 7.70 7.79 6.91 3.95

67 TABLE A.6.5. EFFECTS (%) OF EDUCATION TO UNEMPLOYMENT DURATION OF MEN

(1999-2004)

Men 1999 2000 2001 2002 2003 2004 Copenhagen Frederiksberg 6.51 6.29 4.01 6.38 7.40 6.71 Copenhagen county 5.13 4.92 4.99 5.66 7.47 6.50 Frederiksborg county 4.47 4.33 2.35 5.18 1.74 2.67 Roskilde county 3.27 2.74 7.26 4.59 4.58 4.43 Western Zelland county 4.17 3.47 1.35 4.20 4.01 4.52 Storstrøms county 6.89 3.14 3.49 3.59 4.71 3.28 Funen county 5.70 3.09 3.64 4.43 4.80 4.27 Southern Jutland county 4.71 3.74 3.12 4.65 7.01 3.65 Ribe county 5.16 3.60 4.14 2.87 5.57 3.41 Vejle county 8.21 5.35 4.39 5.62 9.57 6.89 Ringkøbing county 5.73 4.48 4.37 4.58 8.21 5.85 Århus county 6.30 4.98 5.03 5.95 5.42 6.54 Viborg county 6.19 3.49 6.15 3.84 6.09 4.73 Northern Jutland county 5.13 4.66 1.93 5.50 5.87 4.72

TABLE A.6.6. EFFECTS (%) OF EDUCATION TO UNEMPLOYMENT DURATION OF WOMEN

(1999-2004)

Women 1999 2000 2001 2002 2003 2004 Copenhagen Frederiksberg 7.17 7.39 5.97 7.03 8.45 9.36 Copenhagen county 9.23 7.40 4.41 7.10 8.26 9.26 Frederiksborg county 8.67 6.52 8.03 6.46 8.14 5.74 Roskilde county 7.99 8.39 9.02 11.22 7.94 8.59 Western Zelland county 8.15 7.32 4.75 4.18 8.45 10.17 Storstrøms county 9.16 6.85 5.81 5.10 6.09 8.20 Funen county 8.91 8.36 5.62 6.67 8.37 8.25 Southern Jutland county 8.55 6.20 6.55 7.17 9.28 8.40 Ribe county 10.92 5.92 5.98 6.84 8.44 10.38 Vejle county 8.80 8.77 7.96 9.72 12.01 7.45 Ringkøbing county 7.74 8.28 7.50 6.38 8.68 10.98 Århus county 8.71 7.32 7.13 6.64 7.68 8.25 Viborg county 9.63 8.04 5.33 5.78 7.38 9.28 Northern Jutland county 7.40 5.17 5.62 6.25 6.56 7.65

68 TABLE A.6.7. EFFECTS (%) OF OTHER ALMPs TO UNEMPLOYMENT DURATION OF MEN

(1999-2004)

Men 1999 2000 2001 2002 2003 2004 Copenhagen Frederiksberg 2.17 4.43 2.89 0.26 3.73 2.80 Copenhagen county •2.28 2.55 4.39 1.80 4.04 2.45 Frederiksborg county 1.76 0.11 5.93 3.00 1.05 3.62 Roskilde county 0.61 6.68 1.73 2.71 4.99 8.73 Western Zelland county 2.84 •2.85 4.70 0.91 2.82 1.90 Storstrøms county 5.17 0.77 1.65 2.64 4.15 3.48 Funen county •2.95 1.00 0.89 0.85 3.05 4.09 Southern Jutland county 1.63 0.48 3.56 4.10 •1.80 1.54 Ribe county 3.20 2.66 •6.40 •0.17 6.83 5.23 Vejle county 2.73 •0.14 •0.02 3.21 4.73 3.20 Ringkøbing county 5.17 2.66 0.77 0.35 2.77 2.47 Århus county 2.68 0.49 1.48 1.04 4.27 1.86 Viborg county 4.54 2.51 0.81 1.26 •0.10 3.90 Northern Jutland county 2.98 0.30 1.29 1.74 5.39 3.76

TABLE A.6.8. EFFECTS (%) OF OTHER ALMPs TO UNEMPLOYMENT DURATION OF

WOMEN (1999-2004)

Women 1999 2000 2001 2002 2003 2004 Copenhagen Frederiksberg 3.24 4.94 2.78 2.35 6.89 5.18 Copenhagen county 4.90 0.08 0.64 1.91 6.79 5.35 Frederiksborg county •0.76 6.29 •3.80 •0.39 5.91 3.58 Roskilde county 6.00 •3.43 •3.36 •3.47 2.69 •0.81 Western Zelland county 5.91 3.97 0.18 1.43 2.00 •0.57 Storstrøms county 9.22 4.61 0.36 2.81 4.67 4.41 Funen county 3.12 2.02 4.87 1.56 1.84 2.24 Southern Jutland county 8.12 4.40 •0.75 •2.10 2.55 •0.82 Ribe county 1.39 6.55 4.20 •2.71 0.79 3.27 Vejle county 4.95 •0.35 •0.45 1.63 7.08 4.62 Ringkøbing county 6.90 4.80 0.37 3.16 4.84 6.33 Århus county 2.92 1.28 1.36 0.70 5.95 3.36 Viborg county 6.03 5.96 1.46 3.85 5.20 4.97 Northern Jutland county 3.25 6.94 6.63 •0.35 3.74 3.55

69 TABLE A.7.1. UNEMPLOYED MEN AND WOMEN ACTIVATED IN PRIVATE SECTOR

EMPLOYMENT PROGRAMMES IN 1999 –2004, %

1999 2000 2001 2002 2003 2004 2005 Country level Men 5.8 8.6 7.7 19.7 11.8 9.5 9.5 Women 3.2 4.6 4.4 9.7 6.0 5.5 5.3 Copenhagen Frederiksberg Men 3.5 3.8 4.4 4.7 4.4 8 6.6 Women 1.9 2.2 2.6 2.9 3.2 5.5 5.1 Copenhagen county Men 3.3 4.4 5.1 4.9 4.4 9.4 7.2 Women 1.7 3.1 2.8 2.8 2.5 4.3 3.1 Frederiksborg county Men 5 7.8 7.9 17.8 12.3 10.1 7.1 Women 2.6 4.1 4.3 9.2 7.0 4.5 5.2 Roskilde county Men 6.4 7.9 8 20.8 11.6 8.2 14.3 Women 2.3 3.2 4.2 8.4 4.6 3.9 7 Western Zelland county Men 5.5 10.6 7.5 17.5 10.1 9.8 10 Women 3.4 5.6 3 9.5 5.6 4.6 3.6 Storstrøms county Men 5.7 10 10.2 23.2 14.5 11.4 8.8 Women 2.5 3.8 4.7 9.3 7.3 5.5 4.2 Funen county Men 2.9 8.7 9.1 24.5 11.0 10.4 10.2 Women 2.5 3.3 4.5 10.0 5.3 4.8 6.1 Southern Jutland county Men 12.5 16.8 14.9 23.2 15.8 16.1 9.4 Women 5.2 6.2 6.6 9.8 5.7 7.2 4.8 Ribe county Men 4.3 12.8 11 23.4 24.0 14.4 9.6 Women 2.7 4.9 5.2 11.8 9.9 7.1 6.4 Vejle county Men 8 11.6 9.9 24.6 18.8 11.7 10.4 Women 5 7.1 6.2 12.7 9.1 6.9 5.9 Ringkøbing county Men 8.2 11.6 9.6 26.8 15.5 11.5 12 Women 4.3 4.7 5.6 12.3 6.5 7.1 4.9 Århus county Men 8.5 9 6.8 23.2 13.0 8.6 11.8 Women 4 4.5 3.6 11.8 6.1 5.2 5.8 Viborg county Men 3.6 13.9 9.1 22.1 17.0 10.5 11.9 Women 2.7 7.4 5.1 10.9 8.2 7.6 6.2 Northern Jutland county Men 6.7 9.1 8.5 23.3 12.8 11.0 10.3 Women 3 5.9 5.4 10.6 6.1 6.5 6.4

Source: www.jobindsats.dk

70 TABLE A.7.2. UNEMPLOYED MEN AND WOMEN ACTIVATED IN PUBLIC SECTOR

EMPLOYMENT PROGRAMMES IN 1999 –2004, %

1999 2000 2001 2002 2003 2004 2005 Country level Men 11.6 11.5 12 21.5 17.5 13.9 9.3 Women 14.5 15.1 15.8 28.4 20.9 17.5 13.6 Copenhagen Frederiksberg Men 13.1 11.1 10.2 13.5 12 12.8 8.3 Women 14.2 11.2 11 14.3 11.4 15.2 9.4 Copenhagen county Men 16.3 13.5 11.1 13.7 10.5 12.8 8.2 Women 15 13.8 13.5 15.9 12.8 17.3 11 Frederiksborg county Men 12.5 7.5 9.5 18.7 20.1 9.2 6.8 Women 12 10.9 12.5 25.2 25.9 12.7 7.9 Roskilde county Men 7.8 13.5 10.5 16.5 15.2 11.9 7.1 Women 12.9 23.5 18.8 28.5 20.7 20.5 14.3 Western Zelland county Men 12.1 15.9 16.2 18.1 17.5 16.4 13.5 Women 19.4 20.4 19 27.1 19.6 17.8 18 Storstrøms county Men 20.5 16.4 23.4 29.5 27.2 25.5 18.2 Women 21.9 17.8 22.7 34.1 24.4 25.5 21.9 Funen county Men 10.3 13.5 11.4 20.2 15.6 11.3 9.5 Women 15.8 18.1 15.8 24.6 18.8 15.0 12.2 Southern Jutland county Men 16.3 10.4 11.8 25.1 24.8 14.6 11.7 Women 17.3 14.7 16.7 39 31.9 17.8 16.3 Ribe county Men 10.2 15 13.4 24.9 18.7 14.7 9.6 Women 23.7 20.9 26 39 30.7 25.0 16.4 Vejle county Men 7 6.8 7.4 13 11.7 10.2 5.8 Women 10.5 10.8 12.1 23.6 18.0 17.2 11.7 Ringkøbing county Men 8.8 9.3 10 11.4 12.1 9.8 6.9 Women 12.4 14.4 17.9 26.4 20.6 16.3 18 Århus county Men 8.2 6.5 9 21.2 17.2 12.1 8.4 Women 10.7 10 10.5 30.6 22.4 14.6 13 Viborg county Men 8.6 19.8 15.6 22 16.2 13.9 7.3 Women 13.5 21.8 22 27.2 19.6 21.3 12.5 Northern Jutland county Men 10 14.3 18.4 27.8 17.9 17.0 9.6 Women 13.7 19 20.1 33.8 19.4 19.8 15.8

Source: www.jobindsats.dk

71 TABLE A.7.3. UNEMPLOYED MEN AND WOMEN ACTIVATED IN EDUCATION

PROGRAMMES IN 1999 –2004, %

1999 2000 2001 2002 2003 2004 2005 Country level Men 63.8 63.9 57.3 47.7 53.9 48.3 53.8 Women 65.6 65.1 59.6 51.3 58.8 54.7 53.7 Copenhagen Frederiksberg Men 65.3 68.4 57.1 43.9 37.3 56.1 54.4 Women 70 69.5 60.6 46.6 37.4 56.2 57.3 Copenhagen county Men 60.2 65.9 56.1 44.5 37.4 51.7 53.6 Women 71.1 68.2 59.5 48.6 39.1 52.9 57.3 Frederiksborg county Men 68.1 71.1 58.8 54.1 50.7 40.3 44.2 Women 74.8 71.8 60.2 55.2 52.9 48.2 43.8 Roskilde county Men 63.2 71.7 49.3 58.6 62.9 50.0 67.4 Women 65.1 66.7 51.3 59.3 67.5 52.4 68.3 Western Zelland county Men 56.8 59.8 55.7 22.7 44.7 55.1 34.6 Women 60.2 61.9 61.5 28.8 52.5 62.6 35.8 Storstrøms county Men 51.1 46.6 41.9 23 28.7 39.1 28.9 Women 58.6 56.4 56.5 32.8 42.3 50.9 31.5 Funen county Men 83.4 68.3 56.3 52.6 64.4 49.3 68.7 Women 77.1 69.5 59.2 62.5 68.9 55.3 70.1 Southern Jutland county Men 62.3 63.8 59.6 44 49.1 51.5 59.8 Women 69 70.8 64 45.2 52.8 58.4 61.6 Ribe county Men 54.8 54.4 64.4 48.1 51.9 56.1 56.9 Women 44.9 56.6 60.6 45.6 56.2 59.1 46.4 Vejle county Men 62.9 65 48.1 34.4 45.4 40.0 37 Women 60.4 65.1 49.4 38 51.3 46.2 32.9 Ringkøbing county Men 67.7 62.8 53.3 49.1 33.5 34.1 62.9 Women 62.3 62.7 52.8 45.3 40.9 40.0 58.2 Århus county Men 62.5 65 66.5 52.5 63.9 55.3 65.5 Women 65.6 68.1 68 53.7 65.9 60.4 64.3 Viborg county Men 54.4 36.6 45.1 46.6 59.2 44.6 66.8 Women 52.9 44.9 46.3 52.8 66.4 53.8 67.1 Northern Jutland county Men 70.1 66.9 61.3 40.2 62.0 64.0 51.8 Women 71.1 64.5 63.7 50.8 70.5 67.6 48.8

Source: www.jobindsats.dk

72 TABLE A.7.4. UNEMPLOYED MEN AND WOMEN ACTIVATED IN OTHER ALMPs IN 1999 –

2004, %

1999 2000 2001 2002 2003 2004 2005 Country level Men 18.8 16.1 23 11.1 16.9 28.3 27.4 Women 16.7 15.1 20.2 10.6 14.3 22.3 27.4 Copenhagen Frederiksberg Men 18 16.7 28.3 38 46.2 23 30.7 Women 13.9 17.1 25.8 36.2 48.1 23.1 28.2 Copenhagen county Men 20.1 16.2 27.8 36.9 47.8 26.1 31 Women 12.2 15 24.2 32.7 45.6 25.5 28.6 Frederiksborg county Men 14.5 13.6 23.9 9.5 17.0 40.4 41.8 Women 10.6 13.3 23.1 10.5 14.2 34.6 43.1 Roskilde county Men 22.5 6.8 32.2 4.1 10.3 29.9 11.2 Women 19.6 6.5 25.7 3.9 7.1 23.2 10.3 Western Zelland county Men 25.5 13.7 20.5 41.8 27.8 18.7 41.8 Women 16.9 12.1 16.5 34.6 22.3 15.0 42.6 Storstrøms county Men 22.7 27 24.5 24.3 29.6 24.1 44 Women 17.1 22 16.1 23.8 25.9 18.2 42.3 Funen county Men 3.4 9.6 23.2 2.7 9.1 29.0 11.7 Women 4.6 9.1 20.4 2.9 7.1 25.0 11.6 Southern Jutland county Men 8.8 9 13.7 7.8 10.3 17.9 19.2 Women 8.6 8.4 12.7 6 9.7 16.6 17.4 Ribe county Men 30.7 17.8 11.1 3.6 5.4 14.7 23.8 Women 28.8 17.6 8.2 3.5 3.2 8.8 30.8 Vejle county Men 22 16.6 34.6 28 24.1 38.1 46.9 Women 24.2 17.1 32.3 25.8 21.6 29.7 49.4 Ringkøbing county Men 15.3 16.4 27.1 12.7 38.9 44.6 18.2 Women 21 18.3 23.8 16.1 32.0 36.7 18.8 Århus county Men 20.8 19.6 17.7 3.1 6.0 24.0 14.3 Women 19.6 17.4 18 3.9 5.6 19.7 16.8 Viborg county Men 33.4 29.6 30.2 9.3 7.6 31.0 14 Women 31 25.9 26.7 9.1 5.8 17.4 14.2 Northern Jutland county Men 13.2 9.7 11.9 8.6 7.3 8.0 28.3 Women 12.2 10.6 10.8 4.8 4.0 6.1 29

Source: www.jobindsats.dk

73 TABLE A.8.1. UNEMPLOYMENT OF MEN AND WOMEN IN DANISH COUNTIES (EXCL.

BORNHOLM) IN 1999 - 2001, %

1999 2000 2001 Men Women Men Women Men Women Country 4.9 6.5 4.6 6.2 4.5 5.9 Copenhagen Frederiksberg 7.1 6.4 5.8 5.6 5.4 5.2 Copenhagen county 4.1 4.7 4.0 4.5 3.7 4.2 Frederiksborg county 3.4 4.6 3.2 4.2 3.1 4.0 Roskilde county 3.4 4.8 3.2 4.4 3.1 4.1 Western Zelland county 4.8 7.0 4.5 6.8 4.3 6.4 Storstrøms county 6.5 8.4 5.8 7.6 5.6 7.5 Funen county 6.0 7.8 5.7 7.5 5.5 7.0 Southern Jutland county 3.9 7.0 3.9 6.8 3.8 6.4 Ribe county 3.9 6.0 3.5 6.0 3.7 5.8 Vejle county 3.9 6.6 3.7 6.1 4.0 6.3 Ringkøbing county 2.7 5.9 2.7 5.8 2.8 5.6 Århus county 5.5 7.1 5.5 7.0 5.2 6.6 Viborg county 3.4 5.9 3.4 5.9 3.3 5.4 Northern Jutland county 6.1 8.4 6.1 8.4 6.0 8.4

TABLE A.8.2. UNEMPLOYMENT OF MEN AND WOMEN IN DANISH COUNTIES (EXCL.

BORNHOLM) IN 2002 - 2004, %

2002 2003 2004 Men Women Men Women Men Women Country 4.7 5.8 5.7 6.6 5.8 7.0 Copenhagen Frederiksberg 6.0 5.5 7.1 6.3 7.2 6.7 Copenhagen county 4.0 4.1 4.9 4.8 5.2 5.4 Frederiksborg county 3.5 3.9 4.6 4.6 4.4 4.7 Roskilde county 3.5 4.1 4.2 4.7 4.3 5.0 Western Zelland county 4.5 6.0 5.9 7.3 5.9 7.7 Storstrøms county 5.6 7.0 6.4 7.3 6.1 7.2 Funen county 5.5 6.6 6.9 7.7 6.8 8.0 Southern Jutland county 4.3 6.4 5.5 7.7 5.2 7.7 Ribe county 3.7 5.4 4.5 6.2 4.3 6.2 Vejle county 4.0 6.0 4.9 7.0 5.0 7.4 Ringkøbing county 3.0 5.5 3.8 6.3 4.0 6.8 Århus county 5.5 6.5 6.4 7.5 6.5 7.8 Viborg county 3.5 5.3 4.1 5.9 4.2 5.8 Northern Jutland county 5.8 8.1 7.1 8.9 7.8 9.7

74 TABLE A.9.1. FRACTIONS OF LOW-SKILLED WORKERS (1999 - 2001)

1999 2000 2001 Men Women Men Women Men Women Country 41.8 44.8 41.5 45.2 38.6 43.0 Copenhagen Frederiksberg 35.2 40.0 34.6 39.8 32.1 37.7 Copenhagen county 36.3 43.4 35.8 43.1 33.5 40.5 Frederiksborg county 35.8 42.7 35.8 42.9 32.5 40.3 Roskilde county 40.1 46.9 39.4 46.7 36.5 43.9 Western Zelland county 45.1 47.6 44.8 48.0 41.6 45.6 Storstrøms county 43.9 47.9 44.5 48.4 40.7 46.6 Funen county 44.2 45.5 44.1 45.9 41.1 43.4 Southern Jutland county 46.0 47.6 45.5 47.5 42.2 45.5 Ribe county 46.6 45.0 46.6 47.1 43.8 45.2 Vejle county 45.3 47.5 45.4 48.1 42.5 45.7 Ringkøbing county 46.1 47.7 46.1 48.1 43.2 45.9 Århus county 40.5 43.3 40.3 43.7 37.7 41.8 Viborg county 46.7 47.5 46.5 48.5 43.8 47.4 Northern Jutland county 45.4 45.7 45.0 46.8 41.8 44.3

Source: www.statistikbanken.dk

TABLE A.9.2. FRACTIONS OF LOW-SKILLED WORKERS (2002 - 2004)

2002 2003 2004 Men Women Men Women Men Women Country 38.1 42.3 38.0 41.8 34.9 40.4 Copenhagen Frederiksberg 31.2 37.4 31.3 37.1 29.2 35.7 Copenhagen county 33.0 40.1 32.8 39.7 29.8 38.1 Frederiksborg county 32.1 39.2 31.4 38.4 28.6 37.8 Roskilde county 35.8 43.3 35.6 42.9 32.3 41.1 Western Zelland county 41.2 45.3 40.5 44.3 37.1 42.9 Storstrøms county 40.0 45.7 39.6 45.0 36.6 43.3 Funen county 40.7 42.5 40.2 41.9 37.2 41.1 Southern Jutland county 41.7 44.7 41.7 44.4 38.5 42.5 Ribe county 43.1 44.5 43.2 44.2 34.6 42.6 Vejle county 42.3 44.9 42.2 44.1 37.9 42.4 Ringkøbing county 42.5 45.4 42.7 45.0 40.0 43.0 Århus county 37.0 40.5 37.0 41.2 33.8 39.9 Viborg county 43.4 46.1 43.4 43.2 39.7 42.3 Northern Jutland county 42.2 44.3 41.9 44.3 38.3 42.7

Source: www.statistikbanken.dk

75 TABLE A.9.3. FRACTIONS OF WORKERS IN MANUFACTURING (1999 - 2001)

1999 2000 2001 Men Women Men Women Men Women Country 22.5 12.3 22.2 12.1 21.4 11.7 Copenhagen Frederiksberg 10.7 6.9 10.5 6.7 9.8 6.3 Copenhagen county 14.7 9.0 14.2 8.7 13.3 8.2 Frederiksborg county 16.9 10.8 16.7 10.7 16.2 10.4 Roskilde county 17.3 10.0 17.1 9.8 16.4 9.5 Western Zelland county 21.3 11.9 21.4 11.9 21.0 11.6 Storstrøms county 20.6 11.4 20.4 11.2 19.9 10.7 Funen county 26.0 12.1 25.1 11.4 24.0 10.9 Southern Jutland county 29.5 16.1 30.3 16.7 29.3 15.8 Ribe county 28.8 16.9 28.9 16.7 28.4 15.9 Vejle county 30.2 18.1 30.0 17.9 29.3 17.2 Ringkøbing county 33.2 20.0 33.3 19.9 33.2 19.3 Århus county 23.6 11.4 23.1 11.1 22.3 10.9 Viborg county 31.6 18.1 31.6 18.3 31.2 18.3 Northern Jutland county 25.7 12.8 25.2 13.0 24.3 12.5

Source: www.statistikbanken.dk

TABLE A.9.4. FRACTIONS OF WORKERS IN MANUFACTURING (2002 - 2004)

2002 2003 2004 Men Women Men Women Men Women Country 21.3 11.7 20.9 11.3 20.7 11.0 Copenhagen Frederiksberg 9.5 6.2 9.0 6.0 8.6 5.7 Copenhagen county 13.1 8.1 12.6 7.9 12.1 7.7 Frederiksborg county 16.0 10.4 15.7 10.3 15.0 10.2 Roskilde county 16.1 9.5 15.7 9.2 15.5 9.0 Western Zelland county 20.7 11.4 20.2 11.1 20.3 11.1 Storstrøms county 20.3 10.9 19.5 10.6 19.5 10.2 Funen county 23.9 11.0 23.6 10.7 23.5 10.3 Southern Jutland county 29.5 16.3 29.5 15.9 28.4 14.4 Ribe county 28.1 15.8 28.0 15.5 28.5 15.4 Vejle county 29.2 17.0 29.0 16.3 29.0 16.0 Ringkøbing county 34.0 19.6 33.4 18.7 33.3 18.4 Århus county 22.3 10.8 21.5 10.2 21.2 9.9 Viborg county 31.7 18.4 31.5 18.1 31.0 18.0 Northern Jutland county 24.0 12.5 23.8 11.8 23.7 11.8

Source: www.statistikbanken.dk

76 TABLE A.9.5. FRACTIONS OF WORKERS IN PRIVATE SECTOR (1999 - 2001)

1999 2000 2001 Men Women Men Women Men Women Country 74.4 45.1 74.6 45.4 74.8 45.8 Copenhagen Frederiksberg 61.0 46.5 61.7 47.1 63.6 48.8 Copenhagen county 79.0 51.9 79.0 52.3 79.2 52.4 Frederiksborg county 71.3 41.9 71.3 42.1 73.8 43.4 Roskilde county 73.4 38.9 73.4 39.2 72.9 39.6 Western Zelland county 70.4 37.5 70.8 37.3 71.6 38.1 Storstrøms county 68.8 37.3 69.1 37.6 69.3 37.4 Funen county 75.5 42.6 75.9 43.0 75.3 42.8 Southern Jutland county 77.3 47.8 77.9 48.1 77.9 47.7 Ribe county 79.8 48.6 79.9 48.1 80.0 48.1 Vejle county 80.2 48.5 80.3 48.6 80.2 49.0 Ringkøbing county 81.1 50.2 81.5 50.6 82.3 50.8 Århus county 75.7 43.4 76.0 43.7 75.9 44.1 Viborg county 77.8 44.5 78.0 44.8 77.5 45.2 Northern Jutland county 74.9 41.7 74.6 42.0 74.3 41.9

Source: www.statistikbanken.dk

TABLE A.9.6. FRACTIONS OF WORKERS IN PRIVATE SECTOR (2002 - 2004)

2002 2003 2004 Men Women Men Women Men Women Country 75.5 46.0 75.5 45.6 75.6 45.4 Copenhagen Frederiksberg 64.1 49.2 64.6 49.4 64.3 49.0 Copenhagen county 77.4 51.7 77.3 51.6 77.1 50.8 Frederiksborg county 74.4 43.6 73.2 43.3 73.6 43.3 Roskilde county 74.2 40.1 74.3 39.9 74.8 40.4 Western Zelland county 72.9 38.0 73.2 38.1 73.3 38.2 Storstrøms county 71.0 37.6 70.3 37.0 71.3 37.3 Funen county 76.3 42.8 75.7 42.3 76.1 41.6 Southern Jutland county 79.3 47.7 79.1 47.5 79.3 47.1 Ribe county 82.0 48.6 81.6 48.3 81.5 48.1 Vejle county 81.3 49.1 81.3 48.4 81.9 48.4 Ringkøbing county 82.8 51.1 82.9 50.7 82.8 50.2 Århus county 77.3 44.4 77.4 43.7 77.2 43.4 Viborg county 78.3 45.0 78.0 45.0 77.2 45.1 Northern Jutland county 75.4 42.3 75.9 41.4 76.2 41.8

Source: www.statistikbanken.dk

77 TABLE A.9.7. FRACTIONS OF SELF-EMPLOYED WORKERS (1999 - 2001)

1999 2000 2001 Men Women Men Women Men Women Country 10.9 4.0 10.8 4.0 10.7 4.0 Copenhagen Frederiksberg 7.9 3.1 8.2 3.2 8.4 3.3 Copenhagen county 8.5 3.5 8.6 3.5 8.6 3.6 Frederiksborg county 10.4 4.5 10.4 4.5 10.4 4.5 Roskilde county 9.2 3.7 9.4 3.6 9.4 3.6 Western Zelland county 12.1 4.5 12.0 4.5 11.9 4.5 Storstrøms county 13.1 4.8 12.9 4.7 12.8 4.7 Funen county 11.2 4.2 11.2 4.1 11.0 4.1 Southern Jutland county 12.4 4.2 12.3 4.3 12.2 4.2 Ribe county 11.8 4.1 11.6 4.2 11.4 4.1 Vejle county 10.6 4.1 10.4 4.0 10.1 4.0 Ringkøbing county 13.8 4.2 13.4 4.2 13.1 4.1 Århus county 10.4 3.8 10.3 3.9 10.1 3.8 Viborg county 15.1 4.6 14.9 4.6 14.6 4.5 Northern Jutland county 11.9 4.2 11.8 4.2 11.6 4.1

Source: www.statistikbanken.dk

TABLE A.9.8. FRACTIONS OF SELF-EMPLOYED WORKERS (2002 - 2004)

2002 2003 2004 Men Women Men Women Men Women Country 10.6 4.0 10.3 3.8 9.9 3.7 Copenhagen Frederiksberg 8.7 3.5 7.3 3.2 7.3 3.1 Copenhagen county 8.6 3.7 8.2 3.5 8.1 3.5 Frederiksborg county 10.5 4.5 10.3 4.3 10.0 4.3 Roskilde county 9.4 3.7 9.2 3.5 9.0 3.5 Western Zelland county 11.7 4.5 11.4 4.2 11.0 4.1 Storstrøms county 12.7 4.8 12.8 4.6 12.1 4.3 Funen county 10.8 4.1 10.5 4.0 10.0 3.8 Southern Jutland county 12.1 4.1 12.1 4.1 11.5 4.0 Ribe county 11.0 4.1 10.9 3.9 13.6 3.8 Vejle county 9.9 3.9 9.7 3.7 9.3 3.5 Ringkøbing county 12.7 4.0 12.6 3.9 11.8 3.7 Århus county 9.9 3.8 9.6 3.6 9.2 3.5 Viborg county 14.2 4.4 14.1 4.3 13.4 4.1 Northern Jutland county 11.4 4.0 11.2 3.9 10.8 3.8

Source: www.statistikbanken.dk

78 TABLE A.10.1.FRACTIONS OF THOSE OVER 50 AMONG THE 30-59 - AGED UNEMPLOYED

MEN (1999 - 2004)

1999 2000 2001 2002 2003 2004 Country 31.7 32.9 35.0 34.8 34.2 33.9 Copenhagen Frederiksberg 24.0 24.5 24.8 24.7 24.4 24.1 Copenhagen county 33.0 33.3 33.4 33.4 33.2 32.8 Frederiksborg county 35.1 35.3 35.4 35.1 34.6 33.9 Roskilde county 35.8 35.9 35.8 35.2 34.4 33.4 Western Zelland county 32.0 32.8 33.3 33.5 33.5 33.4 Storstrøms county 34.0 34.9 35.6 36.0 36.2 36.2 Funen county 31.1 31.8 32.2 32.5 32.7 32.6 Southern Jutland county 31.8 32.2 32.7 33.0 33.1 33.2 Ribe county 30.3 31.1 31.8 31.9 32.2 32.5 Vejle county 31.3 31.6 31.9 31.9 31.9 31.7 Ringkøbing county 31.1 31.7 32.2 32.5 32.7 32.7 Århus county 30.1 30.9 31.5 31.7 31.9 31.9 Viborg county 29.9 30.8 31.5 32.0 32.4 32.8 Northern Jutland county 30.8 31.6 32.2 32.4 32.6 32.6

Source: www.statistikbanken.dk

TABLE A.10.2. FRACTIONS OF THOSE OVER 50 AMONG THE 30-59 - AGED UNEMPLOYED

WOMEN (1999 - 2004)

1999 2000 2001 2002 2003 2004 Country 30.2 30.6 32.8 32.9 32.3 31.3 Copenhagen Frederiksberg 25.2 25.8 26.1 26.0 25.7 25.3 Copenhagen county 31.2 31.7 32.0 31.8 31.7 31.3 Frederiksborg county 32.3 32.9 33.1 32.9 32.6 32.1 Roskilde county 32.5 33.1 33.4 33.0 32.7 32.0 Western Zelland county 28.7 29.6 30.4 31.0 31.3 31.3 Storstrøms county 30.6 31.8 32.6 33.2 33.8 33.9 Funen county 28.3 29.4 30.2 30.5 30.8 31.0 Southern Jutland county 28.1 29.0 29.8 30.3 30.9 31.3 Ribe county 26.9 27.8 28.6 29.1 29.6 30.1 Vejle county 28.5 29.3 29.8 30.0 30.2 30.1 Ringkøbing county 28.1 29.0 29.7 30.2 30.7 30.9 Århus county 27.4 28.5 29.3 29.7 30.1 30.3 Viborg county 27.2 28.1 29.0 29.4 30.0 30.6 Northern Jutland county 27.9 29.0 29.9 30.4 30.9 31.2

Source: www.statistikbanken.dk

79

Chapter 2

Unemployment, Employment and Inactivity in Denmark: an Analysis of Event History Data

Unemployment, Employment and Inactivity in Denmark: an Analysis of Event History Data

Agne Lauzadyte School of Economics and Management, University of Aarhus

October 6, 2008

Abstract

In this paper I estimate a discrete time hazard model for the exits from the di¤erent labour market states - unemployment, employment and inactivity (or OLF) - in the Danish labour market. I …nd that women and individuals over …fty are more likely to experience long- term unemployment and inactivity. The less educated and unskilled workers are found to be another risk group to face the marginalisation from the labour market. Being previously employed reduces the risk of OLF, and increases the re-entry to employment probability, while long-term unemployment or inactivity makes work- ers more likely to return to these labour market states in the future. Living in the biggest Danish cities, where job competition is high, makes individuals disad- vantaged, but has a positive e¤ect on labour market performance of persons over …fty. And …nally, I …nd that those, who stayed in a job for one year, tend to remain employed, while persons inactive for longer than one year face much higher risk of marginalisation.

Keywords: Discrete time hazard model, Labour market transitions, Marginalisation. JEL-code: C41, J64

83 1. Introduction

This paper examines the ‡ows between the three labour market states - unemployment (U), employment (E) and inactivity (OLF) - in Denmark in the period 1994-2003, and distinguishes the factors having an impact on the transitions above. The goal of the analysis is twofold: …rstly, I capture the phenomenon of repeated unemployment by observing the exits from work to unemployment of previously unemployed individuals, and secondly, I tackle the issue of marginalisation in the labour market1 by examining the risks of leaving a job or unemployment for OLF, and of remaining inactive. Unemployment was high in Denmark during the 1980s and 90s, but since 1994 it decreased signi…cantly partly as a result of the Danish ’Flexicurity’model (see e.g. Andersen & Svarer (2006))2. At the same time, the reform in youth labour market policies (see Jensen et al. (2003)) resulted in a decline in the youth unemployment rate3. A register based sample of the labour market spells of 20-59 years old individuals in a representative 1 per cent sample of the 16-70 year old Danish population in 1994 - 2003 (see Appendix tables A.1.1. - A.1.3.) leads to the result that 42.3% of employment spells of the persons, who in the previous spell were unemployed, end in new unemployment, while those in the previous spell inactive tend to return back into inactivity (47.3% of the spells). Thus, the ’Flexicurity’ model besides the strengths also may have some weaknesses: ‡exible …ring rules and high minimum wages (implied by the generous income transfer schemes) can lead some workers (e.g. least skilled, least educated ones) to become disadvantaged in the labour market4. A number of studies are focused on the transitions between labour market states. For example, Marston (1976) covers three labour market states - employment, unemployment and

1 Marginalisation means high individual rates of inactivity among persons of working age, who are not in education. 2 ’Flexicurity’model consists of three elements: 1) ‡exible hiring and …ring rules (‡ex-element), 2) fairly generous unemployment insurance system (security-element), and 3) Active Labour Market Policies (ALMPs), which are a fairly strict set of rules and regulations regarding availability for work, job search and participation in di¤erent programs. 3 The unemployment rate is de…ned as the proportion of the labour force (i.e. employed plus unemployed persons), which is unemployed, and this de…nition does not cover the individuals outside the labour force. 4 By saying “disadvantaged in the labour market”I mean that the individuals, previously long term unem- ployed and/or inactive, may have shorter job spells and higher risks of repeated unemployment and inactivity compared to the other workers.

84 inactivity - in the US labour market, Meghir & Whitehouse (1997) model the transitions in and out of work for men over the age of 40 in the UK, Nilsen et al. (2000) examine transitions from employment among young Norwegians, while Djurdjevic (2003) tackles the question of re- entry to unemployment by studying the exits from di¤erent employment states and inactivity in Switzerland. Rosholm (2001(2)) applies a three-state competing risks model to analyse marginalisation in the Danish labour market in 1981-1990, when the unemployment rate was mostly high. He covers the ‡ows from unemployment, employment and inactivity of three age groups of Danish youth, and …nds the high youth unemployment rate to be mainly caused by high transition rates into unemployment and inactivity, rather than low transition rates into employment. The present study contributes to the research above by tackling the issue of marginalisation in the Danish labour market further, i.e. by taking into analysis 20-59 years old persons and employing a richer set of explanatory variables representing personal and geographical characteristics of the individuals, and their labour market history. Moreover, I choose the observation period from January 1, 1994 to December 31, 2003 - the time when di¤erent (i.e. much stricter) labour market policies applies in Denmark (see Section 3). I use a longitudinal register-based data set and estimate a discrete time hazard model for the exit from the di¤erent labour market states: unemployment, employment and out of the labour force. The model was introduced by Jenkins (1995) and further developed by Lauer (2003) when she analysed the link between education and risk of becoming unemployed in a French-German comparison. The idea to use this modelling framework in the analysis of re-unemployment was implemented by Djurdjevic (2003) when she analysed the e¤ect of unemployment on the subsequent employment history in Switzerland. Modelling the exit rate from unemployment, employment and inactivity leads to analyse how the transitions out of the states depend on duration. It also leads to analyse, which individuals are more likely to withdraw from the labour market after unemployment. An important issue of the analysis is the experience of women, since they are considered to be a higher risk group. Lauzadyte & Rosholm (2008) found that the unemployment duration is 1.5 - 2 times longer for women than for men. Another question of interest is the impact on labour market experience of the individuals’ age. On the one hand, there is a demand side

85 related danger for elderly workers to be disadvantaged in the labour market; on the other hand there can be a supply side related lack of motivation for those who are close to early retirement. Young individuals, however, also face a higher risk to drop out of the labour force (to complete education or because of personal reasons). It is also important that another - much stricter - labour market policy is applicable to the youth (see Jensen et al. (2003)), while for those over 59 some special rules (in this case - much milder) are valid. Therefore, I exclude from the analysis persons younger than 20 and older than 59, and focus on the transitions from unemployment, employment and OLF based on gender and age. Estimation results show that women and elderly individuals face a higher risk to get outside the labour market and experience higher survival in both unemployment and inactivity. I also …nd non-skilled and low-educated persons, and residents of the biggest Danish cities (except the individuals over …fty), to be disadvantaged in the labour market. Being previously employed reduces the risk of OLF and increases the re-entry to employment probability, while long-term unemployment or inactivity makes workers more likely to get back into these labour market states in the future. Another important …nding is a break in the transition rates from employment and inactivity after 12 months spent there: those who survived in a job one year tend to remain employed, while persons, longer than year inactive, face a much higher risk of marginalisation from the labour market. The structure of the paper is the following. Section 2 gives a brief literature review. Section 3 presents the institutional settings while Section 4 describes and presents the data set used in the study. The modelling framework is explained in Section 5. Section 6 presents and discusses estimation results, and graphs the transitions from the labour market states, based on gender and age, while Section 7 provides the concluding remarks.

2. A Brief Literature Review

While most of the existing research is focused on unemployment, a number of studies examine the other labour market states - employment and inactivity - as well. As a pioneering contribu- tion here I mention Marston (1976), who covers the three labour market states - employment,

86 unemployment and inactivity - in the US labour market. A number of studies on British data have also been carried out. For example, Meghir & Whitehouse (1997) model the transitions in and out of work for men over the age of 40 since 1968, and …nd an increase in earnings to delay job exit, while an increase in social security bene…ts is found to delay return to work. Bradley et al. (2003) run a competing risks model on British Household Panel Survey data of 1992–1997. They identify …ve states - high skilled employment, intermediate skilled employment, low skilled employment, unemployment and OLF - and discover the low skilled workers to be disadvantaged in the British labour market. Cappellari & Jenkins (2004) examine the ‡ows between unemployment, low-paid employ- ment and high-paid employment for British men in 1991–2000, and …nd the low-paid men to be more likely to become unemployed than the high-paid ones (transitions from unemployment to low pay found to be associated with low quali…cations). The goal of the study by Jones et al. (2005) is to evaluate the performance of the National Minimum Wage (MNW) introduced in Britain in 1999, i.e. to …nd out weather this policy measure serves as a stepping stone to higher wages or traps workers in a low wage –no wage cycle. The paper models transitions between di¤erent labour market states – payment at or below the NMW, above the NMW, unemployment and inactivity in 1999-2003 - using the multinomial logit approach and …nd is that for many workers payment at or below the NMW is of relatively short duration, and a substantial number of them move into higher paid jobs. Djurdjevic (2003) estimates a competing risks model to analyse the exits from the di¤erent labour market states - unemployment, employment, characterized by either earning losses, gains or relatively stable earnings, and OLF - in Switzerland (and …nds female, foreign and less skilled workers to experience employment instability), while Nilsen et al. (2000) - examine the transitions from employment among young Norwegians (the individuals with high education, experience, and income are found to have signi…cantly lower probabilities of job exits). Bicakova (2005) focuses on the di¤erences in earnings and labour force status of low-skilled men in France, the UK and the US at the end of the 20th century. She discovers a sizable and signi…cant e¤ect of the wage on the risk of unemployment and inactivity in the UK and the US, but none in France. The low-skilled in France are found to face a relatively similar

87 risk of unemployment, irrespective of their earning capacity, while in the UK and the US, unemployment is found to be concentrated among the low-skilled at the very bottom of the wage distribution. There are several notable Danish contributions within this topic as well. Rosholm (2001(2)) applies a three-state dependent competing risks model to analyse marginalisation in the Danish labour market in 1981-1990, when the unemployment rate was mostly high. He covers the ‡ows from unemployment, employment and inactivity of the three age groups of Danish youth, and …nds the high youth unemployment rate of that period to be caused mainly by high transition rates into unemployment and inactivity, rather than low transition rates into employment. Blume et al. (2007) perform a competing risks analysis of the immigrant-native di¤erence in transition patterns across labour market states based on the Danish register data. They distinguish between three labour market states - wage-employment, self-employment and non- employment (incl. unemployment and OLF) - and …nd that a high proportion of immigrants from non-western countries tend to be marginalised from the labour market relative to natives, and tend to use self-employment to escape marginalisation. The majority of the papers, however, cover the exits from unemployment, and here among others I want to mention van den Berg & van Ours (1999) and van den Berg et al. (2003) for France, Böheim & Taylor (2000) for Britain, Addison & Portugal (2003) for Portugal. A notable Danish contribution is Rosholm (2001(1)), who studies the variance of individual unemployment durations over the business cycle in Denmark. Most of the studies, focused on the labour market transitions, adopt a competing risks model formulation with di¤erent possible destination states (since the factors, that in‡uence the ‡ows to di¤erent destination states, may di¤er) and use a multinomial logit approach (see among others Nilsen et al. (2000), Lauer (2003), Djurdjevic (2003), Jones et al. (2005)).

3. Institutional Settings

The ‘Flexicurity’model – a combination of ‡exible hiring and …ring rules (‡exibility for em- ployers), generous bene…ts for the unemployed (security for employees) and an active labour market policies, which are a fairly strict set of rules and regulations regarding availability for

88 work –is the core of Danish Labour Market Policy. This section gives a short introduction to the policies applicable to unemployed individuals in Denmark. Additionally, I brie‡y introduce the system of retirement and child-related policies, since these may impact the behaviour of older individuals and persons having small children, respectively.

3.1. Unemployment Bene…ts

The unemployment insurance system in Denmark is based on voluntary unemployment in- surance (UI) funds membership, which is open to persons with relevant quali…cations for the speci…c UI-fund, or via regular work within its area. To be entitled to the UI bene…t, an in- dividual must have been employed for at least 52 weeks during the last 3 years and must be a member of UI fund. The UI bene…t cannot exceed 90 % of the last year wage or a given maximum (currently about 22.300 euros annually, taxable income), which is indexed based on the general wage developments. Thus, the replacement rate is highly dependent on previous income, and the 90% compensation in practice is only applicable to the individuals with low wages. When unemployment insurance bene…ts expire, a person is eligible for the social assistance, which is also applicable to the unemployed individuals, who are not members of any UI fund. The social assistance scheme depends on age and marital status of a person, and there are also various means tested supplements. An individual who received the maximum UI bene…t would typically have an income reduction of about 20 –40%. The maximum duration of the UI bene…ts currently (since January 1, 2001) is 4 years, and the entitlement to the bene…ts can only be regained by regular work for at least 6 months within the last 36 months. In contrast to unemployment insurance bene…ts, the social assistance can be received for an unlimited period. The municipalities are responsible for the administration of social assistance and for the organisation of the di¤erent measures helping the social assistance recipients to become self-supporting.

3.2. Labour Market Policies since 1994

A reform on Danish labour Market Policies was carried out in 1994, changing the system from essentially passive income support to a more active approach intended on bringing the unem-

89 ployed individuals back to job. Shortening of UI bene…t period, abolishing the possibility to regain eligibility to UI bene…ts by participation in an ALMP5 and implementation of activation requirements for both UI and social assistance bene…ts recipients were the main elements of the reform. The current active labour market policy is based on the so-called ‘right and duty’principle, which is applicable to both UI and social assistance bene…ts recipients6. The principle is associated with the right of unemployed person to receive the compensation for the loss of income and assistance in the form of an ALMP-o¤er, but also with the duty to participate in ALMPs and other activities when o¤ered to retain eligibility to the UI or social assistance bene…ts.

The length of the period of the UI bene…ts receipt has been reduced signi…cantly since 1994. In 1994, the maximum UI bene…t duration was 7 years, including a 4 year ‘passive’ period and subsequently a 3 year ‘active’period. This duration has gradually been reduced such that currently the maximum UI bene…t period since 2001 is 4 years. After 4 years of UI bene…t receipt, an individual must have at least 26 weeks of full time employment in order to renew bene…t eligibility. Under current rules the passive period lasts 9 months and every unemployed individual, older than 30, who is unemployed for more than 9 months, is required to participate in active labour market programmes. If a person is still unemployed after 26 weeks since programme completion, he is required to participate in another ALMP. Programmes can be o¤ered during the passive period and the unemployed individual has the obligations to accept the o¤ers and to be available for both non-subsidized and subsidized work. However, only a fairly low fraction of the unemployed participate in programmes during the ’passive’period.

3.3. Youth Unemployment Programme

In 1996 another reform was implemented, i.e. the Youth Unemployment Programme was in- troduced, which was directed to unemployed low educated youth. The goal of the Programme

5 The system before the reforms actually implied an in…nite UI bene…t period since despite a formal duration of 7 years, a new bene…t period could be gained by participating in job-training. 6 Initially the ‘right and duty’principle was applicable only to UI bene…ts recipients, but in 1998 it was extended to cover the social assistance recipients as well.

90 was twofold, …rstly, to shorten unemployment duration, and secondly, to foster the economic incentives to undertake education. Mandatory activation of uneducated7 people below 25 after 6 months spent unemployed was introduced by the Programme. The activation could be an education programme of at least 18 months duration, while the social assistance was lowered to the level of study grant8. The Programme was gradually extended to include all young individuals below 309. Cur- rently the social assistance bene…t recipients younger than 30 years have to receive an ALMP- o¤er not later than 3 months after the …rst day on bene…ts. If they do not …nd a job after the end of an ALMP, they have to participate in a new ALMP after 3 months since the end of the previous programme, i.e. they are subject to a more or less continuous treatment in programmes while on bene…ts.

3.4. Policies for Elderly Workers and Retirement

In Denmark, the o¢ cial normal retirement age has been reduced from 67 to 65 years, and currently 65 year old individuals are entitled to a full rate old-age pension. However, the average actual age of retirement is around 61-62 years and the reason of this is partly the voluntary early retirement bene…t programme introduced in 1979, which enables a large part of the labour force to leave the labour market with an income corresponding roughly to unemployment bene…t. The scheme was intended to be an alternative to early retirement for individuals worn down by hard and repetitive physical labour, but instead relatively soon became an important part of the pension system. The age for the early retirement is 60 years, i.e. the residents of Denmark older than 60 are able to retire from a working life before they reach the age of 65. To be able to apply for early retirement bene…t they must have an UI fund membership of at least 25 years. Additionally, individuals over 57 years of age are the group subject to the ’passive’labour market policies.

7 I.e. without any formal education beyond secondary school. 8 This should create an incentive to undertake ordinary education on public study grants or to …nd a job. 9 Those over 24, however, do not face a lower bene…t.

91 3.5. Maternity Leave and Child Allowance

Women with permanent residence in Denmark are entitled to 4 weeks of maternity leave before child birth with a bene…t corresponding to the full salary. After the birth the mother has 14 weeks of maternity leave, while the father is entitled to 2 weeks of paternal leave with the full bene…t during the same period. When the child is 14 weeks old, the parents are entitled to 32 weeks leave with full bene…t to be divided freely between them, which they may choose to spend the leave together or in continuation of each other. The father may begin 32 weeks of leave alongside (or instead of) the mother’s14 weeks. The parents are allowed to prolong their parental leave for up to maximum of 46 weeks from the time when the child is 14 weeks old. But they are still entitled to 32 weeks of leave with the full bene…t, i.e. during the 46 weeks they are paid an amount corresponding to 32 weeks of full bene…t. The bene…t is paid by the municipality or by the employer, who is then reimbursed by the municipality. Additionally, all families liable to pay tax in Denmark receive an allowance for each child below 18. The allowances are independent of the income of the parents and are not taxable. Since 2008 the annual allowance amounts to 16 156 DKK for 0-2 years old children, 12 792 DKK for those between 3-6 and 10 064 DKK for those 7-17 years old.

3.6. Childcare Facilities

There are several childcare alternatives for 0-6-year-old children in Denmark, collectively known as childcare facilities. Most childcare facilities are institutions, but the facilities for childcare in private homes, especially for children under 3 years are also available. The local authorities decide which types of childcare they wish to o¤er the 0-6-year-olds. They are also responsible for establishing, running and supervising the childcare facilities, and may establish them of their own or let private contractors handle it. Municipal childminder arrangements (i.e. the children are cared for in the childminder’s home together with up to 4 other children) and nurseries10 are used for children aged from

10 An institution, where the children are grouped in rooms to which a number of adults are connected, with on average 3 children per childcare employee.

92 6 months to 2-3 years, while the kindergarten is the most common type of childcare for 3-6- year-olds. The kindergartens vary in size, but on average there are 6 children per kindergarten teacher. The parents are also allowed to choose a private childcare instead of using a municipal childcare facility. A private childcare arrangement may have certain advantages, e.g. if the parents work outside the opening hours of the childcare facilities. Rules and subsidy schemes concerning private childcare, however, di¤er among the local authorities.

4. Description of the Data Set

I use the longitudinal register-based data set consisting of event histories for persons belonging to a representative 1 per cent sample of the 16-70 year aged Danish population. The sample is rotating, i.e. it is updated such a way that it is representative in each of the years. The event histories are based on monthly individual unemployment information and on the mandatory pension contributions, made by the employers, and cover the period from January 1, 1985 to December 31, 2003. In the present study I cover the period from January 1, 1994 to December 31, 2003 because better quality data are available after the reform of Danish Active Labour Market Policies introduced in 1994. The data is presented in a person-month format and we can distinguish four states occupied by the individual - employment (E), unemployment (U), recall (or temporary) unemployment (T) and inactivity (or out of the labour force (OLF)). According to UN’sInternational Labour O¢ ce (ILO) de…nition a person is categorised as unemployed if he is out of work, available to work and actively searching for a job11. The unemployment is de…ned as recall unemployment when the unemployed worker returns to the former employer during the …rst three months after becoming unemployed. In this paper I only analyse the unemployment spells that do not end with recall. Recall unemployment and unemployment, shorter than 1 month, is merged with the employment spells. Employed persons are either employees, self-employed or assisting spouses12,

11 All UI bene…ts recipients are classi…ed as unemployed. A fraction of the social assistance recipients (those who are available to work or “employable”) are classi…ed as unemployed, while another fraction (whose “unem- ployable”) belong to OLF category. 12 According to Denmark’sStatistics classi…cation a person is classi…ed as “assisting spouse”if his/her income from assisting spouse is higher than salary.

93 while OLF is the remaining category and includes retirement, maternity leave, education, being a housewife and other non-speci…ed states out of the labour force. Thus, there are three mutually excluding states: Employment (E), Unemployment (U) and Out of the labour force (OLF), and the following transitions are examined: U-E, U-OLF, E-U, E-OLF, OLF-E and OLF-U. I exclude from the analysis individuals younger than 20 and older than 59 and create three age groups - for the persons older than 19 and younger than 30, for the 30-49 years old and for those older than 50 - and allocate them to the groups based on their age at the beginning of the unemployment, employment or inactivity spell.

In the analysis I use a ‡ow sample. For each spell I observe the starting and ending dates, the state occupied and the destination state. To handle the issue of left-censored spells (see e.g. Lancaster (1990) and Steiner (2001)) there is an alternative, namely to use the stock sample, however in this case the model would become very complex, and a number of parameters to be estimated should be reduced. Moreover, this would require a fairly strong stationarity assumption (i.e. the process should be assumed to be constant). Using the ‡ow sample implies conditioning on persons ‡owing into the state. I exclude from the analysis the spells, which were in progress at the beginning of the analysis period, i.e. the individuals who had their unemployment, employment or OLF spell in progress on January 1, 1994 and who never moved out of their labour market state during the period of observation are excluded. Thus, the sample used in this analysis is not a random sample of the population. The older workers could be expected to be a¤ected mostly by selectivity. On the one hand they may enjoy from suitable job and better employment protection, so it could be the case that the older workers included into the sample are a selective group of those with troubles in the labour market. On the other hand, however, the older workers maybe disadvantaged in some situations (like in the case where new working technologies are implemented, requiring new knowledge and/or skills obtained by younger individuals).

94 Table 1. Comparison of individuals included and excluded from the sample, by age

Age, % Unemployment Employment OLF Included* Excluded** Included* Excluded** Included* Excluded** 20-29 35.6 20.4 43.6 25.3 48.1 40.7 30-49 49.8 56.4 45.7 60.8 40.0 44.4 50-59 14.6 23.2 10.7 13.9 11.9 14.9

* Category "included" covers the individuals having U, E or OLF as their …rst spell.

** Category "excluded" covers the individuals who had their U, E or OLF spell in progress on

January 1, 1994 and who never moved out of their labour market state during the period of observation.

The comparison of the workers, included and excluded from the sample13, by age (see Table 1) shows that 23.2 % of the individuals excluded from the sample as unemployed throughout the observation period belong to 50-59 age group in contrast to 14.6 % of those included into the sample, i.e. the older workers maybe disadvantaged in the labour market once unemployed. In the cases of employment and OLF, the fractions of elderly persons among those included and excluded from the sample, however, are rather similar. Table 1 also discovers that 60.8 % of the persons excluded as being constantly employed througout the period of observation belong to the middle age group (compared to 45.7 % of the included ones), and this gives some evidence that the middle-aged persons included into the sample may to some kind be a selective group of those with troubles in the labour market. However, even if this is the case, using the ‡ow sample is still meaningful, since the persons disadvantaged in the labour market is of particular interest. Concerning the 20-29 age group, I …nd that the fraction of its representatives is lower among the individuals excluded from the sample (either as being employed, unemployed or OLF) compared with the included ones. And this is not surprising since young persons are the most ‡exible age group in the labour market.

13 93 persons excluded as unemployed throughout the observation period, 10890 individuals - as employed, while 916 persons - as staying inactive. Thus, the majority of the excluded individuals are employed. Concerning persons included into the sample, I refer to sub-section 3.1.

95 4.1. Sample Composition

As Table 2 indicates, the non-censored unemployment spells in the sample last on average 6.5 months, while the average lengths of the employment and OLF spells are 8.5 and 7 months.

Table 2. Sample composition

U E OLF Number of observations 240985 636133 249172 Number of individuals* 5340 7581 4954 Men 2517 3789 2155 Women 2823 3792 2799 Number of spells 33942 44318 26271 Right-censored spells, % 23.7 44.6 29.7 Non right-censored spells ending in Unemployment, % - 56.3 33.8 Employment, % 79.9 - 66.2 OLF, % 20.1 43.7 - Average length of spell (in months)** 6.5 8.5 7.0

* Number of people having unemployment, employment or OLF as their …rst spell

** Average length of non-censored spell

A large share (80%) of the non-right censored unemployment spells end in employment while the rest end in OLF. Once employed, nearly half (45%) of the individuals tend to remain employed, but on the other hand many individuals move back into unemployment (56% of non- right censored and 31% of all employment spells) or drop out of the labour force. The majority of those, who entered into OLF, tend to get back into employment (66% of non-right censored spells). However, almost 30% of the spells correspond to the individuals remaining out of the labour market. Tables A.2. - A 3. in the Appendix cover the sample split up on gender (Table A.2) and age groups (Table A.3.). Looking at gender di¤erences, I …nd unemployed women to have a higher risk of staying in that position or dropping out of the labour market. One …fth of unemployed men and more than one fourth of unemployed women remain in this state, and only 15% of the non-right censored spells of men, but 25% of women end in the OLF. Once employed, the survival rate here is about 44% for both men and women, but from the transitions out of

96 employment we discover that men are more likely to become unemployed, while women are more likely to leave the labour force. 70% of men, but only 63% of women transit from OLF back into employment. The average duration of non-censored unemployment and OLF spells for men are 1.3 and 0.7 months shorter than for women, while the duration of the employment spell is half a month longer. Di¤erences in the transitions among the age groups are bigger. After becoming employed the youngest age group (20-29 years old people) have lower survival probabilities, and face a higher risk to drop out of the labour market (58% of the non-right censored employment spells). But on the other hand survival in unemployment and in OLF is also lowest for this age group. I discover that only 16% of 20-29 years old unemployed persons stay in that state, while for the elderly (50-59 old) this …gure comes up to 31%. However, the greatest di¤erences exist in the case of OLF. Less than one fourth of the youths, one third of those in the middle age group and more than half of the elderly tend to remain out of the labour market. And 73% of the non-right censored OLF spells in 20-29 age group end in employment, while for 50-59 age group this number comes up to only 53%. Duration of employment spell doesn’tvary between the 20-29 and 30-49 age groups, but is 2 months longer for the 50-59 years old. The unemployment spells, however, are longer for those older than 50 (by 4 months, compared to the middle aged group and by 3 months, compared to the youngest). The non-censored OLF spells are longest for the youths and shortest for those over …fty. This and the fact of the highest survival rates lead to the …nding that once in inactivity, the 50-59 old individuals tend to exit from this state faster than their younger counterparts, but when time spent in OLF increases, they get disadvantaged in the labour market.

4.2. Explanatory Variables

In the analysis I use a number of observable explanatory variables representing personal and geographical characteristics, and labour market history. The three age groups - AGE20-29, AGE30-49 and AGE50-59 - are dummies, which represent the age of the persons in the data set. Female, Married and Immigrant are the dummies for being women, married and immigrants (incl. both 1st and 2nd generation). Mch0_2, Mch3_6, Mch7_17 and Wch0_2, Wch3_6,

97 Wch7_17 are the indicators for the age of the youngest child for men and women; the reference category is having no children. M_Edu11, M_Edu12, M_Edu14, M_Edu16, M_Edu18 and W_Edu11, W_Edu12, W_Edu14, W_Edu16, W_Edu18 represent the lenght of education completed (for men and women) - 10- 11 years, 12 years, 13-14 years, 15-16 years and more than 16 years respectively. The reference here is 9 years or less (primary school education). Experience means work experience of an indi- vidual until the start of the spell, while Previous state is a proxy for labour market attachment, taking the value 1 if the state in the previous spell was employment. To cover the impact of long-term unemployment and inactivity on the future labour market outcomes, I introduce the dummies for being unemployed, employed or inactive longer than six months in the previous spell - PREM _U6, PREM_E6, PREM_OLF6 and PREW _U6, PREW_E6, PREW_OLF6, and PRE5059 _U6, PRE5059_E6, PRE5059_OLF6 - for men, women and 50-59 years old persons respectively. Indicators Copenhagen, Aarhus, Odense and Aalborg indicate place of residence, the ref- erence category being "other place of residence". I also look separately to the e¤ect of place of residence to 50-59 years old individuals: Copenhagen5059, Aarhus5059, Odense5059 and Aalborg5059. I include a set of variables showing the UI fund membership: UI FUND CONSTRUCTION, UI FUND MANUFACTURING, UI FUND TECHNICIANS, UI FUND TRADE, UI FUND CLERICAL, UI FUND ACADEMICS, OTHER UI FUND, and UI FUND SELF-EMPLOYED. Some of UI funds are based on the industry of occupation, while others are based on the edu- cational achvievements of the members. For example, UI FUND MANUFACTURING mainly insures unskilled workers of the manufacturing industry, and UI FUND ACADEMICS covers UI funds, which insure academically educated workers.

5. Methodological Framework

In this paper I estimate a discrete time hazard model for the exit from the di¤erent labour market states: unemployment, employment and out of the labour force. The idea of the hazard rate models (see, for example, Allison (1982), Lancaster (1990)) is to divide the duration spent

98 in a state into a number of time intervals and then to look to each interval whether an individual survived or exited the state. Since the data for this study is available in discrete time intervals - months, I choose a discrete time modelling framework. I distinguish between di¤erent possible destination states and adopt a competing risks for- mulation, since the factors, that in‡uence transitions to di¤erent destination states (for example, transitions from unemployment to employment and from unemployment to OLF), may di¤er, and following the tradition in the latest studies (see among others Nilsen et al. (2000), Lauer (2003), Djurdjevic (2003), Jones et al. (2005)) run a multinomial logit estimation. To examine whether the modelling speci…cation is appropriate I run a couple of speci…cation tests. Firstly I apply the Wald tests for combining the states to test the hypothesis that some of the labour market states could be combined to make the modelling speci…cation to be binomial rather than multinomial. Due to the extremely long computation time of the estimation with the mass points, speci…cation tests are based on estimations with no modelling of unobserved heterogeneity. Further, the MNL requires independence of irrelevant alternatives (IIA): the odds ratio for any pair of choices is assumed independent of any third alternative. Elimination of one of the choices should not change the ratios of probabilities for the remaining choices. Choices that are close, in the sense that their utilities are stochastically correlated, violate the IIA assumption. Therefore I run the Small Hsiao test (Small and Hsiao, 1985) to test the validity of IIA assump- tion. The hypotheses again are tested on the speci…cation without unobserved heterogeneity, i.e. I assume that if the alternatives are independent in this speci…cation, then they are also independent in the less restrictive speci…cation, which allows unobserved heterogeneity. Endogeneity of the past labour market history is another question of interest, that is, the dummy variables for persons being in a particular labour market state for more than 6 months during the previous spell are likely to be endogenous. For example, an individual with a high transition rate from unemployment to job may also experience low transition rate from job to unemployment and OLF due to unobserved reasons, thus he could have a relatively high probability of being employed for a long time during the previous job spell. Estimating the Multivariate Mixed Proportional Hazard (MMPH) model, i.e. the model

99 which allows both for a causal e¤ect and for related unobserved heterogeneity14 could be the way to deal with the issue of endogeneity (see, e.g. Coleman (1990), Rosholm (2001(2)) and van den Berg (2001)). Taking this issue explicitly into account, however, would make modelling fairly complicated and computationally demanding, thus in this study the past labour market outcomes are assumed to be exogenous.

5.1. Model Description

I use the modelling speci…cation introduced by Jenkins (1995) and further developed by Lauer (2003), when she analysed the link between education and risk to become unemployed in a French-German comparison. The idea to use this model in the analysis of re-unemployment was implemented by Djurdjevic (2003), when she analysed the e¤ect of unemployment on the subsequent employment history in Switzerland. Let us assume that T s expresses the time spent by individual i; i 1:::N spent in the ij 2 f g th s s ; s 1:::Si spell of state j; j 1::: before transition to another state or censoring. T 2 f g 2 f g ij s can be partitioned into a discrete number of intervals, It; t 1:::T . In case transition or 2 ij s n o censoring occurs in interval It, we have t = Tij. If a person survives in the state until the s end of interval It, then Tij > t. The set of the observed variables is covered by xij (since the time variation in x may be endogenous, the variables are assumed to be time invariant), while

"ijk represents the unobserved characteristics. The probability that person i moves from the state j to state k; (= j) 1::: in the time interval It (given survival until beginning of It) is 6 2 f g expressed by a destination-speci…c hazard rate and is de…ned as:

s s s s h (t xij;"ijk) = P r(T = t;  = 1 T t; xij;"ijk); (1) ijk j ij ijk j ij > s i = 1; :::; N; t = 1; :::; Tij; j; k = 1; :::; K:

s th Here ijk means the transition indicator, which equals 1 if the s spell of individual i in state j ends in state k and 0 otherwise. Since the exit states are mutually exclusive, the probability

14 Generally, the unobserved determinants of the durations spent in di¤erent states are allowed to be related, and the unobserved determinants of di¤erent durations spent by an individual in the same state are assumed to be identical.

100 th of ending the s spell of state type j for any other state in interval It, can be expressed as:

s s s H (t xij;"ijk) = P r(T = t T t; xij;"ijk) ij j ij j ij >

s = h (t xij;"ijk): (2) ijk j k=j X6 The survivor function shows the unconditional probability that the person stays in the state j until the end of interval It and is de…ned as:

s s S (t xij;"ijk) = P r(T > t xij;"ijk) ij j ij j t s = (1 H (z xij;"ijk)): (3) ij j z=1 Y And …nally, the unconditional probability that individual i moves from his original state j to state k in interval It can be expressed by the product of probabilities that he survives time interval It 1 and that he leaves state j in interval It (given that he had survived until It 1):

s s p (t xij;"ijk) = P r(T = t; k xij;"ijk) ijk j ij j s s = h (t xij;"ijk)S (t 1 xij;"ijk): (4) ijk j ij j

Assuming that all spell observations, conditional on xij(t) and "ijk, are independent, the like- lihood function for the state j can be written as:

s ijk N Si s s s s s j = p (T ) S (T )) ij : (5) L 2 ijk ij 3 ij ij i=1s=1 k=j YY Y6 4 5 s s Here ijk is the transition indicator de…ned above and ij means the censoring indicator, which is equal to 1 if the sth spell of individual i in state j is censored and 0 otherwise (note that s s ij + ijk = 1). k=j 6 s s s NowQ we can introduce indicator yijk, which is equal to 1 when ijk = 1 and t = Tij (see

101 Lauer, 2003) and express the likelihood function in the following way:

s T s 1 yijk N Si ij k=j s s y s P6 j = h (t) ijk 1 h (t) : (6) L ijk 0 ijk 1 i=1s=1k=jt=1 k=j YYY6 Y X6 @ A If we assume the hazard rate to have a multinomial logit form,

s exp jk(t) + jk0 xij + "ijk hijk(t xij;"ijk) = ; (7) j 1 + exp jl(t) + jl0 xij + "ijl l=j P6   equation (6) is a standard multinomial likelihood function, where y represent the transition indicators, and the censored observations enter the likelihood function as an additional state.

The term jk(t) represents the baseline hazard function, which shows the way the hazard rate depends on time. I choose the semi-parametric approach by assuming the baseline hazard function to be piecewise constant (that is jk(t) = jkm, m = 1; :::; Mj , where Mj is the number of intervals for baseline hazard). The following cut-o¤ points for the intervals are used for all hazard rates (the unemployment, employment and OLF spells durations are all measured in months): 3, 6, 9, 12, 15, 18, 21, 24, 36, 60 and 84.

The xij represent the observed variables, which are assumed not to be determined by the future outcomes of the employment, unemployment and inactivity processes.

5.2. Unobserved Heterogeneity

Unobserved heterogeneity "ijk is speci…ed non-parametrically, using the mass point approach

(see Heckman&Singer (1984)). There is assumed a discrete probability distribution for "ijk, i.e. that "ijk can be partitioned into a limited number R of mass points or location parameters

"rjk, r 1:::R , with a given probability P r("rjk). The following conditions are imposed on 2 f g the mass points and their probabilities:

R P r("rjk) = 1 r=1 PR P r("rjk)"rjk = 0 r=1 EP("rjkxij) = 0 Hence, the likelihood function (6) may be rewritten as:

102

s T s 1 yijk R N Si ij k=j s s y s P6 j = P r("rjk) 2 h (t xij;"rjk) ijk 1 h (t xij;"rjk) 3 : L ijk j 0 ijk j 1 r=1 6i=1s=1k=jt=1 k=j 7 X 6YYY6 Y X6 7 6 @ A 7 4 5

Note that the transition rates out of the di¤erent labour market states are estimated sep- arately, and I impose a restriction of no correlation between unobservables in the exits out of di¤erent states, i.e. Corr(vu; ve; volf ) = 0. When modelling transitions out of a given state, I use a “factor loading speci…cation”, imposing perfect correlation between the two unobserved heterogeneity terms. Such parame- terization has been chosen for computational reasons, i.e. to restrict the number of unknown parameters and to limit the computational burden of the estimation of the model.

6. Estimation Results

This section covers the estimation results. The …rst sub-section presents the results of speci- …cation tests, related to the functional form of the hazard rate, while sub-sections 6.2. - 6.4. tackle transitions from unemployment, employment and OLF respectively. Due to the choice of the multinomial logit speci…cation a note on the interpretation of the parameters estimated is needed, i.e. the results presented in Tables 5, 6 and 7 are the parameters, which inform us about the probability of leaving a state for a certain destination state relative to the probability of staying. In other words, I report the probability of leaving for a certain destination state relative to staying in a current state, i.e. the odds ratio: Pk=Pj. Alternatively, the marginal e¤ect of a covariate on the probability of entering state k, i.e. the change in the hazard rate that would result from changing a value of one covariate while keeping other covariates …xed, could be computed, which is not necessarily of the same sign as the parameter involved.

103 6.1. Speci…cation tests

Firstly, I run a series of Wald tests on the (joint) signi…cance of the (sets of) variables and their interactions with the gender and age dummies (see Table 3)15.

Table 3. Wald tests on independent variables

Exits from U Exits from Job Exits from OLF 2 p > 2 2 p > 2 2 p > 2 Tests on coe¢ cients Woman 82.70 0.00 14.86 0.00 16.04 0.00 Immigrant 303.00 0.00 8.27 0.02 70.47 0.00 Age 719.45 0.00 130.11 0.00 404.83 0.00 Place of residence 167.33 0.00 193.51 0.00 41.93 0.00 Experience 149.31 0.00 308.30 0.00 7.86 0.02 Prev. state - employment 4.79 0.04 -- 5.91 0.05 UI fund membership 1396.90 0.00 1412.83 0.00 1661.15 0.00 Tests on interactions Man * Education 131.30 0.00 172.48 0.00 154.16 0.00 Woman * Education 141.75 0.00 24.45 0.01 16.78 0.08 Man * Married 39.56 0.00 6.24 0.04 13.04 0.00 Woman * Married 14.95 0.00 5.61 0.06 2.46 0.29 Man * Children 42.60 0.00 25.57 0.00 6.06 0.42 Woman * Children 478.02 0.00 16.37 0.01 313.41 0.00 Man * Prev. U >6 months -- 129.10 0.00 292.80 0.00 Man * Prev. Job >6 months 37.63 0.00 -- 151.48 0.00 Man * Prev. OLF >6 months 65.75 0.00 167.60 0.00 -- Woman*Prev. U >6 months - - 2.09 0.35 0.90 0.64 Woman*Prev. Job >6 months 6.72 0.04 -- 6.16 0.05 Woman*Prev. OLF >6 months 11.74 0.00 0.87 0.65 - - Age 50-59*Prev. U >6 months -- 12.41 0.00 11.91 0.00 Age 50-59*Prev. Job >6 months 2.69 0.16 - - 304.55 0.00 Age 50-59*Prev. OLF >6 months 1.00 0.61 1.12 0.14 - - Age 50-59*Place of residence 22.69 0.00 9.37 0.31 31.64 0.00 Tests on overall signi…cance (…nally retained speci…cation) Overall Wald test 6583.77 0.00 6238.22 0.00 6481.91 0.00

Bold: signi…cant at 1% level; italic: signi…cant at 10% level

15 In these tests, the null hypothesis that all coe¢ cients associated with given variable(s) or interaction(s) are zeros is tested.

104 The variables and interactions, which proved not to be signi…cant at 10 percent level at least, were dropped from the model. Afterwards, the overall Wald test is used to test (on the basis of the …nal speci…cation with respect to the variables included) the hypothesis that all the slope coe¢ cients of both equations are jointly insigni…cant, which is strongly rejected.

Table 4. Other speci…cation tests

Exit from Unemployment 2 p > 2 Wald test for combining states Combining E and OLF 4653.91 0.00 Combining E and U 7647.81 0.00 Combining OLF and U 2634.82 0.00 Small and Hsiao test for IIA Omitted: E 60.80 0.38 Omitted: OLF 60.90 0.37 Exit from Employment 2 p > 2 Wald test for combining states Combining U and OLF 4527.16 0.00 Combining U and E 9136.23 0.00 Combining OLF and E 7992.39 0.00 Small and Hsiao test for IIA Omitted: U 52.66 0.45 Omitted: OLF 49.31 0.58 Exit from OLF 2 p > 2 Wald test for combining states Combining U and E 4080.94 0.00 Combining U and OLF 6066.27 0.00 Combining E and OLF 6544.96 0.00 Small and Hsiao test for IIA Omitted: U 59.11 0.29 Omitted: E 86.04 0.17

To examine whether the modelling speci…cation is appropriate I use a couple of speci…cation tests (see Table 4). Firstly, I run the Wald tests for combining the states to make the modelling speci…cation to be binomial. That is, I test the null hypothesis that the coe¢ cients of two categories are not signi…cantly di¤erent from each other, and thus that the categories can be

105 collapsed. A series of tests is applied for exits from unemployment, employment and OLF, and the hypothesis is rejected in all the cases (that is, the labour market states can’tbe combined). Furthermore, the Small and Hsiao tests examine the hypothesis of Independence of Irrelevant Alternatives (IIA). If there exists any degree of substitutability among the labour market states, the IIA assumption is violated, and the multinomial logit speci…cation is rejected. The results of the test lead to the …nding that the IIA assumption is supported by the data for all the transitions tested. Thus, the results of both speci…cation tests indicate that the multinomial logit speci…cation is appropriate.

6.2. Transitions from Unemployment

In this sub-section I present and discuss the estimation results of the factors having an impact on the transitions from unemployment to employment and to outside the labour force (the U-E and U-OLF ‡ows). Covering the issue of personal characteristics, it appears that being a woman decreases the chance of leaving unemployment for a job (coe¢ cient of -0.293), but doesn’tin‡uence the probability of getting outside the labour force. Married men are more likely to re-enter employment, once unemployed, and the likelihood of becoming inactive is reduced, while married women face a higher risk of remaining unemployed. Having children a¤ect transitions from unemployment di¤erently for men and women. Women with a baby of two years or younger are less likely to get employed (coe¤. -0.445) and face a higher risk to exit from the labour market (coe¤. 0.481). Children, older than two, but younger than seven also reduce their employment probability (coe¢ cient of -0.175), but have only a slight impact on getting them into inactivity, while those, older than six, lower the chance of exiting unemployment for OLF (coe¤. -0.268). The e¤ect of having a child however is opposite for men - children of all age groups increase their fathers’ employment probabilities, and those, older than two, lower the risk of getting outside the labour force. This di¤erence is not surprising, since women have less time to search for a job and are likely to drop out temporarily from the labour market because of childbearing reasons, while having a family to support may increase the motivation of men to get a job. Concerning the age di¤erences, I …nd the youngest individuals to be the most ‡exible, while

106 those over …fty - to be disadvantaged in the labour market. Compared to the middle-aged (30-49 age group) individuals, the youth are more likely to exit unemployment for both job and OLF (positive coe¢ cients of 0.362 and 0.305 respectively) and these transitions can be a¤ected by the Youth Unemployment Programme (see subsection 3.3.), which creates an incentive either to …nd a job or to start a regular education, i.e. to drop temporary from the labour market. The elderly workers, however, have much lower chances to get employed (coe¤. -0.574) and face comparatively high risk to get into inactivity (coe¤. 0.506), that again can be partially resulted by mild labour market policies for the elderly workers and their behaviour before early retirement (see subsection 3.4.). Another important characteristic is immigration status. The immigrants are found to be a group facing a higher risk to be trapped in this situation - on the one hand they experience a lower chance of leaving unemployment for a job (coe¤. -0.626), but on the other hand they also face a lower risk of getting outside the labour force (coe¤. -0.124). The positive e¤ect of education on employment and the negative e¤ect on OLF suggest that the less educated persons tend to remain unemployed or to withdraw from the labour market, compared to the more educated. Women with 13 -16 years of education, however, are more likely to withdraw from labour market, but the risk is compensated by better job prospects. The unemployment insurance fund membership also plays an important role in explaining the transitions from unemployment. The members of all UI funds face a much lower risk of leaving unemployment for inactivity than the reference category - SID+KAD - which are the two main insurance funds for unskilled men and women. Being previously self-employed or in the trade sector reduces the re-employment probability16. Persons, employed in the previous spell, have a higher likelihood of re-entry to employment (coe¢ cient of 0.365) and a lower risk of inactivity (coe¤. -0.214), while being previously em- ployed more than six months increases the chance of getting a job (coe¤. 0.100) and reduces the risk of inactivity (compared with the re¤erence group - persons, employed less than six months; coe¤. -0.258) for men but is not signi…cant for women.

16 The …ndings above can be partially explained by disadvantage in the labour market of the less educated and unskilled persons. The economic incentives of unemployed individuals may di¤er as well since the replacement rate is highly dependent on previous income, and the UI bene…t of 90% of the previous wage in practice is only applies to the persons with low wages, i.e. less educated and unskilled workers.

107 Table 5. Transitions from Unemployment

Variables To Job To OLF Coe¤. Std. err. Coe¤. Std. err. Female worker -0.293 0.032 0.016 0.051 Age (ref: 30-49) 20-29 years 0.362 0.019 0.305 0.030 50-59 years -0.574 0.039 0.506 0.065 Married man 0.145 0.024 -0.091 0.052 Married woman -0.122 0.032 0.044 0.059 Immigrant -0.626 0.036 -0.124 0.046 Age of youngest child (for women) 0-2 years -0.445 0.030 0.481 0.036 3-6 years -0.175 0.035 0.085 0.050 7-17 years 0.063 0.042 -0.268 0.076 Age of youngest child (for men) 0-2 years 0,152 0,031 -0,070 0,069 3-6 years 0,112 0,039 -0,311 0,102 7-17 years 0,096 0,042 -0,187 0,105 Education, for men (ref:<10 years) 10-11 years 0.005 0.032 0.082 0.058 12 years 0.056 0.036 0.204 0.057 13-14 years 0.117 0.040 -0.111 0.045 15-16 years 0.150 0.022 -0.149 0.084 17-18 years 0.320 0.067 -0.087 0.129 Education, for women (ref:<10 years) 10-11 years 0.095 0.046 -0.126 0.072 12 years 0.321 0.049 -0.124 0.071 13-14 years 0.165 0.035 0.128 0.058 15-16 years 0.347 0.053 0.184 0.049 17-18 years 0.533 0.082 0.195 0.156 Experience 0.014 0.001 -0.017 0.002 Previous state - employment 0.365 0.017 -0.214 0.024 Man prev. employed >6 months 0.100 0.025 -0.258 0.057 Man prev. OLF >6 months -0.334 0.057 0.374 0.070 Woman prev. employed >6 months -0.047 0.038 0.029 0.069 Woman prev. OLF >6 months 0.150 0.078 -0.252 0.091

Bold: signi…cant at 1% level; italic: signi…cant at 10% level

108 Table 5. Transitions from Unemployment (continued)

Variables To Job To OLF Coe¤. Std. err. Coe¤. Std. err. UI fund membership (ref: SID+KAD) Metal 0.085 0.034 -0.694 0.075 Manufact 0.053 0.022 -0.770 0.037 Construct 0.454 0.031 -0.727 0.077 Tech -0.271 0.038 -0.582 0.062 Trade -0.384 0.028 -0.537 0.039 Clerical 0.097 0.030 -0.520 0.049 Acad -0.162 0.043 -0.614 0.074 Uiother -0.031 0.029 -0.558 0.047 Selfs -0.453 0.052 -0.407 0.062 Place of residence (ref: other) Copenhagen -0.220 0.019 0.024 0.031 Aarhus -0.156 0.032 0.207 0.048 Odense -0.121 0.037 -0.072 0.062 Aalborg -0.128 0.038 0.062 0.063 Place of resid. 50-59 years (ref: other) Copenhagen -0.078 0.063 -0.164 0.076 Aarhus -0.145 0.116 -0.382 0.142 Odense -0.176 0.146 0.175 0.159 Aalborg 0.215 0.126 -0.367 0.181 Baseline hazard (ref: 1-3) 4-6 months 0.334 0.018 0.145 0.033 7-9 months 0.103 0.022 -0.082 0.040 10-12 months -0.123 0.027 -0.111 0.045 13-15 months -0.312 0.032 -0.082 0.050 16-18 months -0.321 0.037 0.075 0.053 19-21 months -0.551 0.046 0.040 0.059 22-24 months -0.567 0.052 0.305 0.059 25-36 months -0.685 0.038 0.488 0.041 37-60 months -1.383 0.064 0.255 0.064 61-84 months -1.373 0.078 1.624 0.169 >84 months -2.700 0.166 1.346 0.135 Constant -2.655 0.032 -3.447 0.052 Mass points "1 (Pr ("1) = 0.28) 0.561 0.174 -0.039 0.115 "2 (Pr ("2) = 0.72) -0.218 0.086 0.015 0.191 Bold: signi…cant at 1% level; italic: signi…cant at 10% level

109 The e¤ect of long term inactivity in the previous spell is gender-speci…c: it has a negative e¤ect on the future job prospects (coe¤. -0.334) and increases transitions into OLF (coe¤.0.374) for men. There is an opposite situation for women (the coe¢ cients are 0.150 and -0.252 respec- tively), and this could possibly be explained by the fact that the long-term inactivity of women is often related to child-bearing. The geographical characteristics of unemployed individuals also seem to have an impact on their future labour market prospects. The residents of the counties with the four biggest Danish cities are found to be disadvantaged, compared with the reference category (other place of residence). Living in Copenhagen prolongs a persons’survival in unemployment the most (coe¢ cient to exit for job: -0.220), while the inhabitants of Aarhus county face the highest risk to become inactive (coe¤. to ‡ow into OLF: 0.207). Interesting, however, is the …nding that living in Copenhagen, Aarhus and Aalborg is favourable to the individuals older than …fty - they face a lower risk of getting out of the labour market, once unemployed, while the inhabitants of Aalborg also face better employment prospects (coe¤. 0.215). Turning to the issue of duration dependence, I …nd negative duration dependence in the U-E ‡ows and positive duration dependence while moving from unemployment into inactivity - the long-term unemployment reduces persons’re-employment probability and increases the risk of getting outside the labour market. Finally, I cover the estimations of the individual unobserved heterogeneity. I account for unobserved heterogeneity by employing two mass points of support to improve the model. The presence of these points means that the persons can be divided into two heterogeneous groups. Table 5 indicates that the …rst group of individuals has an above average probability of exiting unemployment for a job (the coe¢ cient is: 0.561). The probabilities for the mass points are: P r("1) = 0:28 and P r("2) = 0:72. This means that for some unobserved reason 28% of unemployed persons are in a better situation of getting back into employment.

To summarise the previous results and to illustrate the duration dependence pattern, I have computed the survivor and hazard functions, based on the gender and the age of individuals. The functions have been calculated for the sub-groups of persons from the estimated coe¢ cients, based on gender and age speci…c characteristics means.

110 Figure 1: Survival in Unemployment

1.A.:

Survival in Unemployment, by Gender 1 .8 .6 .4 Estimated survivor functions .2 0 10 20 30 40 Time in months

Men Women

1.B.:

Survival in Unemployment, by Age 1 .8 .6 .4 .2 Estimated survivor functions 0

0 10 20 30 40 Time in months

20•29 Age group 30•49 Age group 50•59 Age group

111 Figure 2: Transitions from Unemployment to Employment

2.A.:

Transition from Unemployment into Employment, by Gender .15 .1 .05 Estimated hazard functions hazard Estimated 0

0 10 20 30 40 Time in months

Men Women

2.B.:

Transition from Unemployment into Employment, by Age .15 .1 .05 Estimated hazard functions 0

0 10 20 30 40 Time in months

20•29 Age group 30•49 Age group 50•59 Age group

112 Figure 3: Transitions from Unemployment to OLF

3.A.:

Transition from Unemployment into OLF, by Gender .04 .03 .02 Estimated hazard functions .01 0 10 20 30 40 Time in months

Men Women

3.B.:

Transition from Unemployment into OLF, by Age .05 .03 Estimated hazard functions .01 0 10 20 30 40 Time in months

20•29 Age group 30•49 Age group 50•59 Age group

113 The graphs show a steady decline over time in the survivor functions for both genders and all age groups. It turns out that women remain in unemployment longer than men. Di¤erences in the survival among the age groups, however, are sharper. The elderly individuals experience higher survival rates, while the youngest group has the best chances to leave unemployment (after two years of unemployment 18% of the youth, but 49% of the elderly remain in such a situation). Transitions to employment have a spike in 4-6 months of unemployment spell and then decline gradually, but remain rather stable after the …rst year on bene…ts. The ‡ows into OLF, on the contrary, become stable after 6 months of unemployment, but increase sharply after 21 months. Concerning the gender issue, I …nd women to experience lower transitions to a job (especially in the …rst year of unemployment) and higher ‡ows to inactivity, while the age-based analysis discovers the youngest to have highest chances to move to a job, but also to OLF. The elderly persons are less likely to transit to a job, and the middle-aged have the lowest risk of becoming inactive.

6.3. Transitions from Employment

This sub-section tackles transitions from employment into unemployment and inactivity (the E-U and E-OLF ‡ows). Concerning the e¤ect of gender and personal characteristics, I …nd that being a woman or an immigrant increases the risk of re-entry to unemployment (the coe¢ cients are 0.202 and 0.158 respectively), but doesn’tin‡uence the ‡ows into inactivity. Being married plays a positive role for the employment situation (the probabilities of getting unemployed (for men) or inactive (for women) are reduced). Children have di¤erent impact for men and women. The e¤ect of having a child, younger than seven, is found not to be signi…cant regarding exits from employment of women17. Children of seven or older, however, make their mothers more attached to the labour market (the coe¤. for exits from employment is: (-0.193) + (-0.231) = (-0.424). For men, though, children of all age groups make them more likely to stay in job. And this could again be explained by the positive e¤ect of having family to support to their motivation of remaining employed.

17 The well developed childcare facilities (see subsection 3.5.) allows Danish women to keep a full time job, and their labour force participation eventually approached that of the men.

114 Table 6. Transitions from Employment

Variables To U To OLF Coe¤. Std. err. Coe¤. Std. err. Female worker 0.202 0.050 -0.020 0.069 Age (ref: 30-49) 20-29 years -0.255 0.030 0.672 0.041 50-59 years 0.398 0.048 0.348 0.085 Married man -0.107 0.039 -0.005 0.069 Married woman 0.010 0.052 -0.237 0.083 Immigrant 0.158 0.054 0.043 0.068 Age of youngest child (for women) 0-2 years 0.035 0.052 0.029 0.066 3-6 years 0.055 0.053 -0.043 0.075 7-17 years -0.193 0.069 -0.231 0.105 Age of youngest child (for men) 0-2 years -0.012 0.050 -0.360 0.089 3-6 years -0.131 0.068 -0.410 0.132 7-17 years 0.006 0.072 -0.269 0.154 Education, for men (ref:<10 years) 10-11 years -0.039 0.049 -0.165 0.072 12 years -0.306 0.055 0.152 0.059 13-14 years -0.176 0.037 -0.285 0.062 15-16 years -0.245 0.069 -0.152 0.090 17-18 years -0.726 0.123 -0.541 0.143 Education, for women (ref:<10 years) 10-11 years -0.073 0.071 -0.102 0.097 12 years -0.129 0.076 0.011 0.079 13-14 years -0.160 0.057 -0.043 0.087 15-16 years -0.325 0.094 0.169 0.116 17-18 years 0.151 0.148 0.135 0.188 Experience -0.025 0.002 -0.037 0.003 Man prev. U >6 months 0.472 0.053 -0.292 0.028 Man prev. OLF >6 months -0.398 0.032 0.762 0.051 50-59 years - prev. U >6 months -0.066 0.076 0.510 0.153

Bold: signi…cant at 1% level; italic: signi…cant at 10% level

115 Table 6. Transitions from Employment (continued)

Variables To U To OLF Coe¤. Std. err. Coe¤. Std. err. UI fund membership (ref: SID+KAD) Metal 0.482 0.057 -1.231 0.114 Manufact 0.677 0.034 -1.152 0.061 Construct 0.698 0.052 -1.246 0.115 Tech 0.211 0.063 -1.017 0.101 Trade 0.261 0.045 -1.164 0.073 Clerical 0.162 0.052 -1.088 0.075 Acad 0.185 0.078 -0.939 0.096 Uiother 0.476 0.046 -0.835 0.075 Selfs -0.366 0.092 -0.989 0.124 Place of residence (ref: other) Copenhagen -0.179 0.030 0.215 0.032 Aarhus -0.152 0.050 0.457 0.045 Odense 0.081 0.054 0.135 0.062 Aalborg 0.232 0.055 0.245 0.068 Baseline hazard (ref: 1-3) 4-6 months -0.282 0.028 -0.042 0.035 7-9 months -0.566 0.036 -0.436 0.046 10-12 months -1.138 0.051 -0.892 0.061 13-15 months -1.983 0.081 -1.579 0.089 16-18 months -1.982 0.084 -1.637 0.096 19-21 months -2.027 0.090 -1.797 0.108 22-24 months -2.111 0.097 -1.276 0.088 25-36 months -2.401 0.063 -1.883 0.069 37-60 months -3.041 0.072 -2.166 0.069 61-84 months -3.473 0.104 -2.363 0.091 >84 months -3.490 0.111 -2.218 0.092 Constant -2.395 0.050 -2.663 0.067 Mass points "1 (Pr ("1) = 0.24) -0.405 0.262 -0.314 0.243 "2 (Pr ("2) = 0.76) 0.128 0.043 0.099 0.026

Bold: signi…cant at 1% level; italic: signi…cant at 10% level

116 Concerning the age, belonging to 20-29 age group reduces the risk of unemployment (coe¤.: - 0.255), but the probability of dropping outside the labour market increases18. Thus, the overall risk of exiting employment is higher (the coe¤. is: (-0.255) + 0.672 = 0.417. The elderly persons, however, face the highest risk of exiting job for both unemployment and inactivity (coe¤.: 0.398 + 0.348 = 0.746). The education level is an important factor, helping to remain in a job once employed. Men in all educational groups are found to be in a favourable employment situation, compared to the reference group (those with nine or less years of education), and the most educated individuals face the lowest risk of leaving a job (the coe¢ cients for the hazards to unemployment and OLF for the group with 17-18 years of education are -0.726 and -0.541 respectively). For women, the years of education have no impact to the transitions from job to inactivity, but lower the risk of unemployment (coe¤. for the group with 15-16 years of education is -0.325)19. The e¤ect of previous job experience on the transitions from employment is found not to be strong, though signi…cant, while previous unemployment or inactivity, longer than 6 months, makes men more likely to get trapped back into these stages (the coe¢ cients for the exits to unemployment and OLF are 0.472 and 0.762). Long-term unemployment in the previous spell gives a signi…cant coe¢ cient of 0.510 for the ‡ow into inactivity of 50-59 years old individuals. Unemployment insurance fund membership has a complementary role in explaining the ‡ows from employment. The self-employed individuals are most likely to remain employed and face the lowest transitions into both unemployment and OLF (the coe¤.: -0.366 and - 0.989). It is surprising, that members of all other UI funds experience higher risk of re-entry to unemployment than the reference category - members of SID+KAD funds. But on the other hand, they also experience a much lower risk of moving outside the labour market, and thus are in a better employment situation than the unskilled workers. Concerning the place of residence factor, I …nd the residents of Copenhagen, Aarhus, Odense and Aalborg to be more likely to exit employment than the reference category - residents of other places.

18 And this can be explained by the fact that many young workers leave their jobs for regular education, which is covered by OLF category. 19 These, again, can be explained by the better job prospective for educated individuals and by the economic incentives to stay employed, since in the case of job loss income replacement rate is comparatively lower for the higher wage earners.

117 Figure 4: Survival in Employment

4.A.:

Survival in Employment, by Gender .9 .8 .7 Estimated survivor functions .6 0 10 20 30 40 Time in months

Men Women

4.B.:

Survival in Employment, by Age 1 .9 .8 .7 Estimated survivor functions .6 0 10 20 30 40 Time in months

20•29 Age group 30•49 Age group 50•59 Age group

118 Figure 5: Transitions from Employment to Unemployment

5.A.:

Transition from Employment into Unemployment, by Gender .08 .06 .04 .02 Estimated hazard functions 0

0 10 20 30 40 Time in months

Men Women

5.B.:

Transition from Employment into Unemployment, by Age .1 .08 .06 .04 .02 Estimated hazard functions hazard Estimated 0

0 10 20 30 40 Time in months

20•29 Age group 30•49 Age group 50•59 Age group

119 Figure 6: Transitions from Employment to OLF

6.A.:

Transition from Employment into OLF, by Gender .05 .04 .03 .02 Estimated hazard functions Estimated .01 0

0 10 20 30 40 Time in months

Men Women

6.B.:

Transition from Employment into OLF, by Age .05 .04 .03 .02 .01 Estimated hazard functions 0

0 10 20 30 40 Time in months

20•29 Age group 30•49 Age group 50•59 Age group

120 Living in Odense or Aalborg increases the ‡ows from employment to unemployment and OLF. The inhabitants of Copenhagen and Aarhus experience lower transitions to unemploy- ment, but are at a higher risk of moving into inactivity. Thus, the e¤ect of place of residence to the exit from employment is slightly positive (coe¤.: (-0.179) + 0.215 = 0.036) for Copenhagen and positive (coe¤.: (-0.152) + 0.457 = 0.305) for Aarhus. There is negative duration dependence in both E-U and E-OLF ‡ows. The probability of exiting a job declines with the time spent employed, and there is a sharp decline in the baseline hazards after the …rst year of employment. And lastly, concerning the unobserved heterogeneity issue, I …nd that for some unobserved reasons 76% of individuals face a higher risk to exit a job once employed (coe¤. is: 0.128 + 0.099 = 0.227).

The age and gender speci…c survivor and hazard functions illustrate the …ndings above. It turns out that men stay employed longer than women, but the gender speci…c survival di¤erences are slight. However, looking to the E-U and E-OLF ‡ows (especially during the …rst year of employment), there is evidence that men are more likely to leave a job for unemployment, while women more tend to move into inactivity. There is a minor di¤erence in job survival of the youth and the elderly individuals, and both age groups survive as employed shorter than their middle-aged counterparts. But here, again, there is a di¤erence in the transitions - the youngest persons are much more likely to drop out of the labour market, while those older than 50 face the risk of re-entry to unemployment. Another interesting and important …nding is a sharp decline in the transition rates into unemployment and inactivity after the …rst year of employment: the individuals, who survived employed one year, tend to remain in that state.

6.4. Transitions from OLF

The analysis of the transitions from OLF (the OLF-E and OLF-U ‡ows) leads to the result that being a woman reduces the chance of exiting OLF for employment (coe¢ cient of -0.209). Being married has negative impact on the ‡ows from inactivity for men, but the in‡uence of marriage to the exit for employment is not signi…cant. Having small children makes women less likely to leave OLF for job (coe¤.: -0.674 and -0.148). Immigrants face a higher overall probability of leaving OLF. But this increase is due to the

121 exit from OLF to unemployment (the coe¢ cient for the OLF-E ‡ow is -0.182), thus they are at the risk of being trapped in a long-term unemployment. The age factor tends to play an extremely important role in explaining the transitions from inactivity. The youngest are found to be in the most favourable situation (coe¢ cient for the OLF-E ‡ow is 0.424), but here I want to pay attention to the elderly workers, who are found to be strongly disadvantaged on the labour market, compared to their younger counterparts. Being older than 50 sharply lowers the chances of exiting for both employment and unemployment, and the coe¢ cient for overall OLF exit probability is (- 1.574). The number of years of education has a positive in‡uence on the exit from OLF for employ- ment. Men with 17-18 years and women with 15-16 years of education face the most favourable re-employment prospective (coe¤. 0.351 and 0.158). The members of all UI funds survive shorter in inactivity than the reference category. Thus, the unskilled workers are found to be another problematic group, facing the risk of marginalisation from the labour market20. Those, who were previously employed, experience higher transitions to re-employment (co- e¢ cient of 0.552) and have lower chance to leave inactivity for unemployment (coe¤. -0.715). This is in line with previous …ndings that inactive persons, who were previously employed, tend to return to employment, while those previously unemployed face a risk of re-entry to unem- ployment (see Appendix table A.1.3.) Being previously employed longer than six months gives better job prospective for both men and women (the coe¢ cients for the OLF - E ‡ow are 0.246 and 0.189 respectively), while being previously long-term unemployed increases the risk of the OLF-UI ‡ow for men. The past employment history plays an extremely important role for the 50-59 years old per- sons - those with the previous job spell longer than 6 months are more likely to leave inactivity for both new job or UI. Individuals having more than six months of past unemployment history, however, face a much higher risk of staying trapped out of the labour market, compared to the reference group - persons, unemployed less than six months (the coe¢ cient for the exit from OLF is -1.524 + (-1.443) = (-2.967)).

20 Like in the case of unemployment, these …ndings can be explained by disadvantage in the labour market of the less educated and unskilled persons and by the economic incentives of the individuals.

122 Table 7. Transitions from OLF Variables To Job To U Coe¤. Std. err. Coe¤. Std. err. Female worker -0.209 0.048 -0.021 0.050 Age (ref: 30-49) 20-29 years 0.424 0.029 0.042 0.030 50-59 years -1.116 0.085 -0.458 0.078 Married man 0.002 0.027 -0.143 0.028 Immigrant -0.182 0.050 0.326 0.045 Age of youngest child (for women) 0-2 years -0.674 0.039 -0.025 0.036 3-6 years -0.148 0.048 0.075 0.049 7-17 years 0.098 0.065 0.039 0.073 Education, for men (ref:<10 years) 10-11 years 0.061 0.049 0.001 0.059 12 years 0.068 0.041 -0.562 0.059 13-14 years 0.129 0.042 -0.154 0.047 15-16 years 0.083 0.058 -0.443 0.079 17-18 years 0.351 0.094 -0.623 0.131 Education, for women (ref:<10 years) 10-11 years 0.119 0.066 -0.049 0.073 12 years 0.128 0.055 0.076 0.073 13-14 years -0.027 0.058 0.053 0.061 15-16 years 0.158 0.075 0.005 0.095 17-18 years 0.041 0.123 0.160 0.158 Experience 0.005 0.002 -0.005 0.003 Previous state - employment 0.552 0.024 -0.715 0.026 Man prev. empl. >6 months 0.246 0.030 -0.684 0.053 Man prev. OLF >6 months -0.593 0.040 0.306 0.030 Woman prev. employed >6 months 0.189 0.068 0.032 0.041 50-59 years - prev. empl. >6 months 0.194 0.088 0.274 0.102 50-59 years - prev. U >6 months -1.524 0.183 -1.443 0.094

Bold: signi…cant at 1% level; italic: signi…cant at 10% level

123 Table 7. Transitions from OLF (continued)

Variables To Employment To UI Coe¤. Std. err. Coe¤. Std. err. UI fund membership (ref: SID+KAD) Metal 0.371 0.066 1.085 0.073 Manufact 0.301 0.038 1.032 0.036 Construct 0.431 0.068 0.760 0.081 Tech 0.192 0.064 1.040 0.059 Trade 0.063 0.044 0.919 0.040 Clerical 0.511 0.043 0.811 0.051 Acad 0.219 0.066 1.334 0.070 Uiother 0.267 0.050 0.906 0.048 Selfs 0.169 0.074 0.239 0.090 Place of residence (ref: other) Copenhagen 0.027 0.014 -0.087 0.030 Aarhus -0.078 0.033 -0.192 0.047 Odense -0.040 0.044 -0.125 0.059 Aalborg -0.136 0.049 0.016 0.057 Place of resid. 50-59 years (ref: other) Copenhagen 0.300 0.100 0.298 0.096 Aarhus 0.545 0.213 0.588 0.181 Odense 0.051 0.236 0.225 0.197 Aalborg 0.059 0.270 0.238 0.209 Baseline hazard (ref: 1-3) 4-6 months 0.414 0.024 0.076 0.027 7-9 months 0.125 0.030 -0.361 0.038 10-12 months 0.518 0.031 -0.117 0.043 13-15 months -0.701 0.059 -1.223 0.082 16-18 months -0.526 0.060 -0.971 0.081 19-21 months -0.774 0.073 -1.243 0.101 22-24 months -0.480 0.070 -1.032 0.102 25-36 months -0.994 0.055 -1.548 0.079 37-60 months -1.767 0.075 -2.226 0.098 61-84 months -2.459 0.123 -3.590 0.220 >84 months -3.508 0.206 -6.442 1.000 Constant -3.066 0.047 -2.943 0.050 Mass points "1 (Pr ("1) = 0.86) 0.131 0.114 -0.093 0.309 "2 (Pr ("2) = 0.14) -0.805 0.175 0.571 0.068 Bold: signi…cant at 1% level; italic: signi…cant at 10% level

124 Concerning the geographical pattern, again I …nd the residents of counties with the four biggest Danish cities to be disadvantaged in the labour market. Living in Copenhagen, however, slightly increases the chance of getting back into employment (coe¤.: 0.027), but on the other hand slightly decreases the transitions into unemployment (coe¤.: -0.087). The negative e¤ect of residing in Aarhus is not very strong as well (coe¤.: -0.078), but here I want to mention that the inhabitants of Aarhus and Odense are less likely to leave OLF for unemployment, i.e. there is a danger for these individuals to be discouraged to search for a job and thus, to be marginalised and remain out of the labour market. The residents of Aalborg face the lowest transitions from OLF to employment. The inhabitants of Copenhagen and Aarhus, older than …fty, are found to be in a favourable labour market situation, compared to the reference group - the coe¢ cients for their exits from inactivity are 0.300 + 0.298 = 0.598 and 0.545 + 0.588 = 1.133 respectively. As in the case of the U-E ‡ow, I here discover the presence of negative duration dependence. In the transitions from OLF into employment, however, I …nd individuals more likely to move for a job during the …rst year (especially in 10-12 months) in OLF. But the situation changes drastically after the …rst year of inactivity. For the OLF-U transitions, I observe negative duration dependence during the whole period of inactivity, i.e. the time, spent outside the labour market, discourages people to search for a job. And …nally, concerning the individual unobserved heterogeneity, I discover 14% of inactive persons to have a lower chance of leaving OLF for a job.

Now, again, I illustrate the …ndings above by graphing gender and age based survivors and hazards. I observe women to survive in OLF longer than men. Concerning the age of the persons the youngest are least likely to remain inactive while the elderly are found to be the risk group. After two years in OLF 30% of 20-29 old individuals, but 75% of those, older than 50, remain in this state. There are no gender di¤erences in the transitions from OLF into unemployment, but women experience lower chances of leaving inactivity for a job, especially during the …rst year outside the labour market. Youth are more likely to move into job than their older counterparts, while looking to the OLF-U ‡ow, I …nd the middle-aged to be in the best situation. In both transitions again I …nd the age based di¤erences to be mostly expressed in the …rst year of inactivity.

125 Figure 7: Survival in OLF

7.A.:

Survival in OLF, by Gender 1 .8 .6 Estimated survivor functions survivor Estimated .4 0 10 20 30 40 Time in months

Men Women

7.B.:

Survival in OLF, by Age 1 .8 .6 .4 Estimated survivor functions survivor Estimated .2 0 10 20 30 40 Time in months

20•29 Age group 30•49 Age group 50•59 Age group

126 Figure 8: Transitions from OLF to Employment

8.A.:

Transition from OLF into Employment, by Gender .1 .05 Estimated hazard functions 0

0 10 20 30 40 Time in months

Men Women

8.B.:

Transition from OLF into Employment, by Age .15 .1 .05 Estimated hazard functions Estimated 0

0 10 20 30 40 Time in months

20•29 Age group 30•49 Age group 50•59 Age group

127 Figure 9: Transitions from OLF to Unemployment

9.A.:

Transition from OLF into Unemployment, by Gender .06 .04 .02 Estimated hazard functions 0

0 10 20 30 40 Time in months

Men Women

9.B.:

Transition from OLF into Unemployment, by Age .03 .02 .01 Estimated hazard functions hazard Estimated 0

0 10 20 30 40 Time in months

20•29 Age group 30•49 Age group 50•59 Age group

128 And the last and very important …nding is a sharp decline in both transitions after the …rst year in OLF. In the previous chapter I mentioned such break to be existent in the case of employment, however here it is even more expressed. For example, during the 10-12 months of inactivity the hazards for the 20-29, 30-49 and 50-59 age groups are 13%, 10% and 4%, but after one year, spent in OLF, they drop to 4%, 3% and 1% respectively. Similar behaviour is observed in gender based hazard rates, and in the hazards from OLF into unemployment.

7. Conclusions

In this paper I use the longitudinal register-based data and estimate a discrete time hazard model for the exits from the di¤erent labour market states - unemployment, employment and inactivity - in the Danish labour market. I distinguish between the di¤erent possible destination states, adopt a competing risks formulation and run multinomial logit estimation. The estimations results …nd women to face a higher risk of getting outside the labour market and to experience higher survivals in both unemployment and inactivity. The youth is found to be the most ‡exible age group, most likely to leave employment and unemployment for OLF, but also having the highest transitions back into employment, while those over …fty are found to face the highest risk of long-term unemployment and marginalisation from the labour market. Years of education make persons less likely to exit for unemployment or inactivity, and help them to …nd a job once unemployed or outside the labour force, while the unskilled workers are found to be the highest risk group to drop out of the labour force, and to remain in that state. Being previously employed reduces the risk of OLF, and increases the re-entry to employ- ment probability, while long-term unemployment or inactivity makes workers more likely to get back into these labour market states in the future. Living in the biggest Danish cities, where job competition is high, makes individuals disadvantaged, but has a positive e¤ect on labour market performance of persons over …fty. Finally, I …nd a break in the transition rates from employment and inactivity after 12 months spent there: those, who survived in a job one year, tend to remain employed, while persons, longer than one year inactive, face much higher risk of marginalisation from the labour market.

129 References

[1] Addison, J. T. and Portugal, P. (2003). "Six Ways to Leave Unemployment", IZA Discussion Paper No. 954.

[2] Allison, P. (1982). "Discrete-time Methods for the Analysis of Event Histories", In S. Leinhardt (Ed.), Sociological Methodology 1982 (p. 61-97). San Fransisco: Jossey-Bass.

[3] Andersen, T.M., Svarer, M. (2006). "Flexicurity - den danska arbetsmarknadmodellen", Ekonomisk Debatt, nr. 1, s. 27-39.

[4] Bicakova A., (2005). "Unemployment vs. inactivity: an analysis of the earnings and labor force status of prime age men in France, the UK, and the US at the End of the 20th Century", mimeograph, Department of economics Johns Hopkins University, February.

[5] Blume, K., Ejrnæs, M., Nielsen, H. S., Würtz, A. (2007). "Labour Market Transitions of Immigrants with Emphasis on Marginalization and Self-Employment", Working paper, AKF, Institute of Local Government Studies.

[6] Boheim, R., Taylor, M. (2000). "Unemployment duration and exit states in Britain", Institute for Social and Economic Research, Working Paper 2000-1, University of Essex.

[7] Bradley, S., Crouchley, R., Oskrochi, R. (2003). "Social exclusion and labour market transitions, a multi-state multi-spell analysis using the BHPS", Labour Economics, Vol. 10, No. 6, 2003, 659-679.

[8] Cappellari, L., Jenkins, S. (2004). "Modelling Low Income Transitions", Journal of Applied Econometrics, Vol. 19, No. 5, 2004, pp. 593-610.

[9] Coleman, T.S., (1990). "Unemployment behaviour: evidence from the CPS work experi- ence survey", in Y. Weiss and G. Fishelson, editors, Advances in the theory and measurement of unemployment, Macmillan, Basingstoke.

[10] Djurdjevic, D., (2003). "Unemployment and Under-employment: The Case of Switzer- land", University of St. Gallen Economics Working Paper No. 2003-18.

130 [11] Hausman, J., McFadden, D. (1984). "Speci…cation Tests for the Multinomial Logit Model", Econometrica, 52(5), 1219-1240.

[12] Heckman, J.J., Singer, B. (1984). "A Method of Minimizing the Impact of Distributional Assumptions in Econometric Models for Duration Data", Econometrica, 52, 271-320.

[13] Jenkins, S.P., (1995). “Practitioner’s corner – Easy estimation methods for discrete- time duration models”, Oxford Bulletin of Economics and Statistics, 57(1), 129-138.

[14] Jensen, P., Rosholm, M., Svarer, M. (2003). “The Response of Youth Unemployment to Bene…ts, Incentives, and Sanctions”, European Journal of Political Economy, 19, 301-316.

[15] Jones, M. K., Jones, R. J., Murphy, P. D., Sloane, P. J. (2005). "The Dynamics of the National Minimum Wage: Transitions Between Di¤erent Labour Market States", IZA Discussion Paper No. 1690.

[16] Lancaster, T. (1990). "The Econometric Analysis of Transition Data", Cambridge University Press, Cambridge.

[17] Lauer, C. (2003). "Education and Unemployment: A French-German Comparison", Discussion Paper 03-34, ZEW.

[18] Lauzadyte, A., Rosholm, M. (2008). "County, Gender and Year Speci…c E¤ects of Active Labour Market Programmes in Denmark", Working paper, The University of Aarhus.

[19] Marston, S.T., (1976). “Employment Instability and High Unemployment Rates”, Brookings Papers on Economic Activity,1, 169-203.

[20] Meghir, C., Whitehouse, E. (1999). "Labor Market Transitions and Retirement of Men in the UK." Journal ofEconometrics, September 1999, 79(2),pp. 324-54.

[21] Nielsen, O.A., Risa, A.E., Torstensen, A. (2000). “Transitions from Employment among Young Norwegian Workers”, Journal of Population Economics, 13, 21-34.

[22] Rosholm, M. (2001(1)). "Cyclical variations in unemployment duration", Journal of Population Economics, 14, 173-191.

131 [23] Rosholm, M. (2001(2)). "An Analysis of the Processes of Labour Market Exclusion and (Re-) Inclusion", IZA Discussion Paper No. 332.

[24] Small, K. A. and Hsiao, C. (1985). "Multinomial Logit Speci…cation Tests", Interna- tional Economic Review, 26 (3), 619-627.

[25] Steiner, V. (2001). "Unemployment persistence in the West German labour market: negative duration dependence or sorting?", Oxford Bulletin of Economics and Statistics, 63 (1), 91-113.

[26] van den Berg, G. J., van Ours, J.C.: (1999). "Duration Dependence and. Heterogeneity in French Youth Unemployment Durations", Journal of Population Economics, 12.

[27] van den Berg, G. J. (2001). "Duration Models: Speci…cation, Identi…cation, and Mul- tiple Durations", Handbook of Econometrics, Vol. 5, Chapter 55, Elsevier.

[28] van den Berg, G. J., van Lomwel, A. G. C., van Ours, J. C. (2003 ). "Nonparamet- ric Estimation of a Dependent Competing Risks Model for Unemployment Durations", IZA Discussion Paper No 898.

132 APPENDIX

TABLE A.1.1.: EMPLOYMENT SPELLS IN 1% REPRESENTATIVE SAMPLE (1993•2003)

If previous spell • If previous spell • unemployment OLF Spells ending in unemployment, % 42.32 14.21 Spells ending in OLF, % 9.15 47.27 Right•censored spells, % 48.52 38.52 Non right•censored spells ending in: Unemployment, % 82.22 23.12 OLF, % 17.78 76.88 Average lenght of spell (in months) 12.70 16.88

* I use the longitudinal register•based data set of event histories for persons belonging to a representative 1% sample of the 16•70 year aged Danish population. Persons younger than 20 and older than 59 have been excluded.

TABLE A.1.2.: UNEMPLOYMENT SPELLS IN 1% REPRESENTATIVE SAMPLE (1993•2003)

If previous spell • If previous spell • employment OLF Spells ending in employment, % 66.41 45.27 Spells ending in OLF, % 12.38 26.16 Right•censored spells, % 21.20 28.56 Non right•censored spells ending in: Employment, % 84.29 63.38 OLF, % 15.71 36.62 Average lenght of spell (in months) 7.03 7.37

* I use the longitudinal register•based data set of event histories for persons belonging to a representative 1% sample of the 16•70 year aged Danish population. Persons younger than 20 and older than 59 have been excluded.

133 TABLE A.1.3.: OLF SPELLS IN 1% REPRESENTATIVE SAMPLE (1993•2003)

If previous spell • If previous spell • employment unemployment Spells ending in employment, % 57.99 22.95 Spells ending in unemployment, % 13.83 44.11 Right•censored spells, % 28.18 32.95 Non right•censored spells ending in: Employment, % 80.74 34.22 Unemployment, % 19.26 65.78 Average lenght of spell (in months) 9.81 8.81

* I use the longitudinal register•based data set of event histories for persons belonging to a representative 1% sample of the 16•70 year aged Danish population. Persons younger than 20 and older than 59 have been excluded.

134 TABLE A.2.: SAMPLE COMPOSITION (GENDER BASED)*

Unemployment Employment OLF

Men Number of observations 106749 331236 104834 Number of individuals** 2517 3789 2155 Number of spells 16510 22214 11321 Non right•censored spell ending in Unemployment, % • 60.3 29.9 Employment, % 84.9 • 70.1 OLF, % 15.1 39.7 • Lenght of non•censored spell (in months) 5.9 8.6 6.6 Women Number of observations 134236 304897 144338 Number of individuals** 2823 3792 2799 Number of spells 17432 22104 14950 Right•censored spells, % 27.2 44.8 31.1 Non right•censored spell ending in Unemployment, % • 52.2 36.8 Employment, % 74.7 • 63.2 OLF, % 25.3 47.8 • Lenght of non•censored spell (in months) 7.2 8.3 7.3

* I use the longitudinal register•based data set of event histories for persons belonging to a representative 1% sample of the 16•70 year aged Danish population. Persons younger than 20 and older than 59 have been excluded. ** Number of people having unemployment, employment or OLF as their first spell

135 TABLE A.3.: SAMPLE COMPOSITION (AGE GROUPS)*

Unemployment Employment OLF 20•29 group Number of observations 70486 257835 100399 Number of spells 12072 19341 12637 Right•censored spells, % 15.9 38.0 22.5 Non right•censored spell ending in Unemployment, % • 41.8 26.9 Employment, % 77.7 • 73.1 OLF, % 22.3 58.2 • Lenght of non•censored spell (in months) 5.5 8.2 7.7 30•49 group Number of observations 120201 306605 94326 Number of spells 16914 20230 10510 Right•censored spells, % 27.0 50.5 31.9 Non right•censored spell ending in Unemployment, % • 68.6 40.4 Employment, % 83.4 • 59.6 OLF, % 16.6 31.4 • Lenght of non•censored spell (in months) 6.5 8.2 6.3 50•59 group Number of observations 50298 71693 54447 Number of spells 4956 4747 3124 Right•censored spells, % 31.3 46.1 51.6 Non right•censored spell ending in Unemployment, % • 75.7 46.8 Employment, % 74.0 • 53.2 OLF, % 26.0 24.3 • Lenght of non•censored spell (in months) 9.6 10.2 5.5 * I use the longitudinal register•based data set of event histories for persons belonging to a representative 1% sample of the 16•70 year aged Danish population. Persons younger than 20 and older than 59 have been excluded.

136

Chapter 3

Optimal Introduction Time and Length of Active Labour Market Programmes in Denmark

Optimal Introduction Time and Length of Active Labour Market Programmes in Denmark

Agne Lauzadyte School of Economics and Management, University of Aarhus

October 6, 2008

Abstract

In this study, I estimate the e¤ects of ALMPs on UI bene…t recipients in Den- mark. Firstly, I examine the e¤ects of the programmes, depending on the time spent in unemployment before entry into the programme. Then I analyse how the performance of the two types of ALMPs - Education and Other ALMPs - varies with the length of the programmes, and lastly, I carry out the programme entry time dependent analysis of short-term Education and Other ALMPs. I use the timing-of-events model introduced by Abbring and van den Berg (2003) and …nd that only Private Job training, and short-term Education and the residual ALMPs reduce unemployment duration. The results in this paper are in line with the …ndings of studies in other OECD countries, i.e., in favour of activation of unemployed persons in their …rst year of unemployment. However, the results do not support activation in the …rst 1-6 months of UI bene…t spells. ALMPs in Denmark are found to be ine¤ective after two years on UI bene…ts.

Keywords: Active labour market policy, Early activation, Programme length, Timing-of- events, Duration model. JEL-code: C41, J64

141 1. Introduction

A number of studies in the OECD countries focus on the main goal of Active Labour Market Programmes (ALMPs) - reintegrating the long-term unemployed into the labour market. While most of the papers estimate the e¤ectiveness of ALMPs, assuming that all individuals respond in the same way to a given programme, there is a growing interest in models allowing for hetero- geneous programme e¤ects. For example, Weber & Hofer (2004(b)) investigate the dependence of programme e¤ect on varying entry times for a low cost job-search assistance programme in Austria, while Flores-Lagunes et al. (2007) estimate the e¤ects of length of exposure to the Job Corps (JC) training programme in the US labour market. In this study, I estimate the e¤ects of the Danish ALMPs on unemployment insurance (UI) bene…t recipients. The goal of the paper is twofold: …rstly, I examine the e¤ects of the programmes, depending on the time spent in unemployment before entry into the programme and secondly, I examine how the performance of the two types of ALMPs - Education and Other ALMPs - varies with the length of the programmes. The unemployment rate was high in Denmark during the 1980s and in beginning of the 1990s, reaching a peak in the year 1994. In the mid-1990, a comprehensive reform of Active Labour Market Policies (ALMPs) was enacted. A shift in policy from mostly passive cash support to activation of unemployed individuals to upgrade their quali…cations was implemented by the reform, and since then the use of active labour market programmes has been increased. The passive period with receipt of UI bene…ts has gradually been reduced by the reform from 4 years in 1994 to 1 year in 2001. In the so-called active period with UI bene…ts an individual is obliged to programme participation 75% of the time. Since 2007 the policy has become tightened, and every unemployed individual, older than 30, who is unemployed for more than 9 months, now has to be activated. If a person is still unemployed after two and a half years, he is obliged to full time participation in ALMPs during the remaining bene…t period up to maximum 4 years. Programmes can also be o¤ered during the passive period, in which case the unemployed individual has to accept them. However, only a fairly low fraction of the unemployed participate in programmes during the passive period (see Table 1).

142 Table 1. Activations in Danish ALMPs (January 1, 1999 - November 15, 2005)*

Age group 1-6** 7-12** 13-18** 19-24** >24** Ratio A1 30-39 5.5 24.5 34.4 35.4 33.6 40-49 5.8 23.2 35.3 37.0 33.3 50-59 5.9 21.2 33.1 34.2 31.3 Ratio B2 30-39 36.5 20.4 18.5 11.1 13.5 40-49 34.5 20.2 19.3 11.9 14.0 50-59 30.4 19.7 20.2 12.4 17.3

* Register data based calculations

** Months of UI bene…ts spell

The comprehensive use of ALMPs is often mentioned as one of the explanations of the favourable development in the Danish labour market, but it is a contentious issue whether this was the major reason for the decline in the unemployment rate. Most analyses on Danish data have shown that ALMPs in general have quite disappointing e¤ects, see e.g. Rosholm & Svarer (2004) and Munch & Skipper (2004). Most of the studies seem to agree, however, that a program of subsidized employment in the private sector shortens unemployment duration; see Bolvig et al. (2003), Graversen (2004) and Rosholm & Lauzadyte (2008). There is an ongoing debate in the OECD countries on the e¤ectiveness regarding reaching the main goal of the programmes - reintegrating the long-term unemployed into the labour market - and the interest in early ALMP activations is growing. However, the main problem regarding early entry into the ALMPs is increasing costs of the programmes, since the target group of participants increases. Furthermore, there is the risk that the persons who participate in the ALMPs would have found a job anyway i.e. the deadweight costs will become higher. According to the budget for 2007, 15.8 million DKK are set aside for the Active Labour Market Measures (for both UI and the social assistance recipients), so the optimal allocation of the

1 Ratio between the number of persons activated in ALMPs in each interval of the UI bene…t spell and the number of unemployed persons at the start of each interval. 2 Ratio between the number of persons activated in ALMPs in each interval of the UI bene…t spell and the number of persons activated in ALMPs in January 1, 1999 - November 15, 2005.

143 means is crucial. For example, for the period under study 5.5-6 % of 30-59 years old unemployed individuals participated in ALMPs in the …rst 6 months of their unemployment spell (see Table 1), but about 30-35% of all activations in the period belonged to these early activations. Studies in other OECD countries …nd early ALMP interventions to be e¤ective, but conclude that there is no need for very early activations; see e.g. Weber & Hofer (2004(a)), Weber & Hofer (2004(b)) and Carling & Larsson (2005). In an experiment based analysis on Danish data, Rosholm (2007) discovers the intensi…cation of labour market policy at a very early stage of unemployment to be highly e¤ective, but the e¤ects are found to be associated with the threat e¤ect of having to enter the programme; not because of the speci…c treatments. This study is a contribution to the discussion above. I use the timing-of-events model introduced by Abbring and van den Berg (2003) and estimate two separate e¤ects of the ALMPs in a duration model: a locking-in e¤ect and a post-programme e¤ect. In the empirical studies, it is often found that the exit rate from unemployment to employment is relatively low during the programme participation period, and this decrease in the exit rate has been termed a locking-in e¤ect. The reason for the decrease in job search intensity may be that there is less leisure time for making job applications, or the participant might also want to complete the education programme o¤ered. If this is the case, the locking-in e¤ect by itself increases the duration of the spell with UI bene…ts. The post-programme e¤ect covers the period after participation in a programme. If a person experiences higher exit rate to a job, compared to the period before programme participation, there is a positive post-programme e¤ect, i.e. the programme reduces UI bene…t spell duration. Intuitively, the locking-in e¤ect is estimated by comparing the UI-to-job transition rate of programme participants with the persons not participating in a programme. While estimating the post-programme e¤ect, the UI-to-job transition rates of programme participants are com- pared with the individuals who have not participated in a programme. These comparisons of transition rates are made at equal durations of the UI spell. To avoid selection bias in the estimates of the locking-in and the post-programme e¤ects, the model accounts for possible selectivity in the in‡ow into programmes. Finally, I calculate the net e¤ects of ALMPs on unemployment duration. The results of the optimal timing of ALMPs in this paper are in line with the …ndings of

144 studies in other OECD countries, i.e. in favour of activation of unemployed persons in their …rst year of unemployment. However, the results do not support activation in the …rst 1-6 months of UI bene…t spells. The programme length based analysis discovers that only the short-term programmes of up to 6 weeks reduce unemployment duration. The structure of the paper is the following: the next section gives a brief literature review. Section 3 describes the Danish labour market and brie‡y presents an experiment done in two Danish counties. Section 4 is devoted to a presentation of the data set, while section 5 explains the econometric model. Estimation results are discussed in section 6, and section 7 concludes.

2. A Brief Literature Review

While most of the studies focus on the main goal of ALMPs - reintegrating the long-term unemployed into the labour market - there is a growing interest in the prevention of long-term unemployment, with activation taking place in the early stages of unemployment. There is, however, little empirical evidence on the question how the e¤ectiveness of the programmes varies with the timing of programme entry. However, a few studies on the question above exist. For example, Blundell et al. (2004) evaluate the labour market impact of the British New Deal for Young People3. The early entry into the programme appears to have caused an increase in the probability of young men (who have been unemployed for at least six months) …nding a job within the next four months. The estimated increase is about 5 percentage points, and 1 percentage point of the 5 percentage points is found to occur due to the Gateway services, such as job search assistance (rather than the wage subsidy element). Further, the treatment impact is found to be much larger in the …rst quarter of introduction compared to the subsequent two quarters. Brodaty et al. (2002) examine the relationship between the time spent unemployed and the e¤ects of the youth programmes in France in 1986-1988 and 1995-1998. They estimate a competing-risks duration model to derive the propensity scores that are used to construct the matching estimates of the programme e¤ects. They …nd the youth employment programmes to

3 The New Deal is a compulsory programme a¤ecting all young people claiming unemployment bene…t for at least six months. The programme o¤ers a combination of treatments, particularly job assistance for four months and a wage subsidy paid to employers.

145 be less e¤ective in 1995-1998 than ten years before, while on the whole they discover the active labour market policies in France to be bene…cial to the long-term unemployed young workers as well. Weber & Hofer (2004(b)) investigate the dependence of the programme e¤ect on varying entry times for a low cost job-search assistance programme in Austria. They measure the programme e¤ect by a shift in the transition rate into employment upon programme entry, using the timing-of-events method, and …nd the programme e¤ect to be positive and not to vary signi…cantly for programme entry times during the …rst year of unemployment, but to drop drastically thereafter. Carling & Larsson (2005) measure the e¤ect of Utvecklingsgarantin (UVG) or early inter- vention for Swedish youth, with the goal of preventing unemployment spells longer than 100 days4. They found no evidence that the programme signi…cantly improves the future labour market situation of the youth and concluded that early intervention in the unemployment spell was not important. Most of the studies on the optimal timing of ALMP participation, however, conclude in favour of early activation of unemployed individuals, on the other hand, they …nd no evidence supporting the importance of very early programme entry. There exists some literature on the e¤ects of the ALMPs, depending on the length of the programme. For example, Flores-Lagunes et al. (2007) estimate the e¤ects of length of exposure to the Job Corps (JC) training programme5 in the US labour market. They measure average causal e¤ects of di¤erent lengths of exposure to JC using the “generalized propensity score” under the assumption that the length of the individual’s JC spell is randomly assigned. They …nd the estimated (marginal) e¤ects of an additional week of training to decline with the total length of enrolment in the programme. In addition, they …nd the hispanics to bene…t more from the programme, compared to whites and blacks, especially when programme duration is longer. Mealli et al. (1996) run a competing risks model to estimate the e¤ects of the Youth Training

4 By guaranteeing the assignment to a programme within 100 days, long-term open unemployment would be avoided. On the other hand, such a guarantee might provide an attractive alternative to regular employment, and thereby extend the time youths stay detached from ordinary working life. 5 The Job Corps is US’largest job training programme for disadvantaged youth. Average programme duration is 28 weeks.

146 Scheme (YTS) in the UK labour market, while Upward (2002) employs the matching technique to discover the outcomes of the same programme. Both studies …nd the full-term YTS spells6 to be associated with signi…cantly higher employment probabilities than the spells that were interrupted earlier. Torp (1994) examines the impacts of training programmes in the Norwegian labour market and …nds the employment e¤ects to vary with the duration of training and from one …eld of training to another. The interesting …nding is that in general short courses (5- I0 weeks) and long courses (more than 30 weeks) tend to perform better than medium-long ones. In this paper, I follow the recent tradition of programme evaluation and use the timing- of-events model for identifying treatment e¤ects in a duration model framework, developed by Abbring & van den Berg (2003), to estimate the e¤ects of ALMPs in Denmark depending on the time of programme entry. For education and other ALMPs I also estimate the e¤ects depending on program length.

3. The Danish Labour Market

The unemployment bene…t system in Denmark consists of two elements: unemployment in- surance (UI) bene…ts and social assistance bene…ts. Only the members of UI funds who have been employed for at least 52 weeks within the last three years are eligible for UI bene…ts. Un- employed persons not belonging to this group (approximately 20% of the labour force) receive unemployment assistance administrated by municipalities. Since this paper is focused on UI recipients, I cover only the policies that apply to individuals receiving UI bene…ts. The Danish labour market policy is characterised by the so-called ’Flexicurity’model, con- sisting of three elements: 1) ‡exible hiring and …ring rules (‡ex-element), 2) a fairly generous unemployment insurance system (security-element), and 3) the "rights and obligations" prin- ciple. The principle guarantees an individual the right to compensation for the loss of income, but also places on him an obligation to take active steps to get back into employment. On the one hand, the society has the obligation to help the individual to improve his situation, but on the other hand society also has the right to make requirements of the individual concerned.

6 Normally, the Youth Training Scheme is limited to 2 years

147 The ALMPs are classi…ed into 4 types by the National Labour Market Authority: * Subsidized employment programmes with private employers. The individual is employed in the private sector for a 6 - 9 months period, and the employer is paid the subsidy, corresponding to roughly 50% of the minimum wage. * Subsidized employment programs with public employers. These programmes o¤er the individual temporary (6 - 12 months) jobs in public sector institutions. * Education/training programmes. These include all types (usually short-lasting) of training programmes, tailored to the background of the unemployed individual concerned. * Other programmes, which include all programmes that cannot be classi…ed within one of the categories above. A variety of programmes is covered by this residual group, for example job search assistance, competence detection programmes, individual specialized job training (in case the unemployed individual cannot participate in ordinary training programmes), etc.

3.1. Labour Market Policy since 1994

The length of the period during which an unemployed individual can receive UI bene…ts has been reduced signi…cantly by the reform of active labour market policies, introduced in 1994. Before the reform, the maximum UI bene…t duration was 7 years, including a 4 year ‘passive’ period and subsequently a 3 year so-called ‘active’period. This duration has gradually been reduced to 4 years in 2001 (consisting of a 1 year ’passive’and 3 years ’active’period). During the active period, the unemployed person is required to participate in active labour market programmes and the time span between programme participation has to be shorter than 6 months7. Programmes can also be o¤ered during the passive period based on the regional labour market council’sevaluation of the regional needs, or in order to test the availability for work of a certain individual or group of individuals. The unemployed individual has the obliga- tion to accept all programmes o¤ered and to be available for both non-subsidized and subsidized work. However, only a fairly low fraction of the unemployed participate in programmes during the initial passive period. Before the reform in 1994, participation in programmes led to renewed eligibility for UI

7 Until 2003, persons who had been unemployed for 1 year should participate in ALMPs for at least 80% of their remaining time in unemployment.

148 bene…ts. In 1994, this option for renewal of eligibility was abandoned, such that only ordinary full time employment of at least 26 weeks during the past 52 weeks would lead to the renewed eligibility.

3.2. An Experiment in Two Danish Counties

A "natural experiment" in labour market policy was carried out in Storstrøm and South Jutland counties in Denmark in the winter of 2005 - 2006. Half of the newly unemployed UI bene…t recipients (who registered themselves as unemployed during the period from November 1, 2005 to February 28, 2006) were assigned to participate in the experiment, while the remaining individuals belonged to the control group. Selection into the treatment and control groups was based on birth dates of the persons within a given month, thus the experiment was truly random. The intensi…cation of labour market policies involving information, early compulsory par- ticipation in job search assistance programmes, frequent meetings with employment o¢ cers and full-time programme participation for at least three months for people who had not found a job within 18 weeks from becoming unemployed were the key elements of the experimental treatment. Persons in the control group had to participate in the programmes after 1 year of unemployment. Job search assistance and monitoring were less intensive for the control group. The experiment based analysis (see Rosholm (2007)) leads to interesting results; …rst, the intensi…cation of labour market policy is highly e¤ective, leading to increases in the exit rate from unemployment ranging from 20-50%, varying by region and elapsed unemployment du- ration. Secondly, despite the overall positive e¤ects of the experiment, none of the speci…c treatments could be demonstrated to have a positive e¤ect. Thus, a possible explanation of these …ndings is that the e¤ects of the programmes occurred because of the intensi…cation of the labour market policy regime (i.e. due to the threat e¤ect), and not because of the speci…c treatments.

3.3. Danish Labour Market Policies since 2007

Since 2007 policies have been tightened further, and every unemployed individual, older than 30, who has been unemployed for more than 9 months, now has to be activated. If a person is

149 still unemployed after two and a half years, he has the obligation of full time participation in ALMPs during the remaining duration of the UI bene…t spell.

4. Data Set

I use a data extracted from an event history data set developed by the Danish National Labour Market Authority (NLMA). The event histories are based on the administrative registers, which record and govern the payments of public income transfers, as well as the register in which the employment agencies record the unemployeds’ participation in ALMPs. Using these event histories constructed by the NLMA itself, the employment agencies determine the risk that an individual becomes long-term unemployed (Hammer et al., 2004), so in this respect not only the underlying information, but also the event histories themselves, are considered to be a very reliable data source. The data used in this paper covers the period from January 1, 1999 to November 15, 2005. The records are updated on a weekly basis and include all spells where the unemployed has received a public income transfer. Since this data is used for administrative purposes, it is not merged with other registers containing information on such variables as education and work experience. In this paper I concentrate on the unemployment spells of workers who are eligible for UI bene…ts since the information available for UI recipients is of a much higher quality than for social assistance recipients. An unemployment spell is de…ned as the period in which an indi- vidual is either openly unemployed or participates in an ALMP. If a person has four consecutive weeks out of open unemployment where he does not receive any other public income transfer, then he is treated as having found a job. If an individual has more than four weeks out of unemployment where he receives other transfers, the unemployment spell is characterised as right censored. Temporary lay-o¤ unemployment is eliminated by excluding from our samples all unemploy- ment spells lasting less than four weeks (note: about 40% of the unemployment spells belong to temporary unemployment, more than 90% of them last less than four weeks; see Jensen et al. (1990)) since the ALMPs are not used in the case of a short-term unemployment. Periods out

150 of unemployment, shorter than four weeks, which usually include paternity leave (two weeks), holiday periods (typically up to three weeks), or short periods of sickness bene…t payments, are excluded from the unemployment spell. If an unemployed individual starts on his second programme during the same unemployment spell, the unemployment spell is treated as right censored at the time at which the second programme period starts. This is done in order to avoid having to model selection problems in connection with sequences of programme participation. I create three data sets: for persons older than 29 and younger than 40, for the 40-49 - old, and for those 50-59 years old. The samples consist of 5 % of the observations drawn randomly, i.e. 128518 observations (30-39 age group), 91303 observations (40-49 age group) and 80899 observations (50-59 age group).

4.1. Explanatory Variables

In the analysis, I use a number of observable explanatory variables representing personal char- acteristics and labour market history. The indicators of gender and marital status and two indicators representing the country of origin have been included (the reference category is na- tive Danes). The indicator that the individual lives alone is SINGLE. WOMAN is a dummy of being woman. IMMIGRANT FROM DC covers the …rst and second generation immigrants from developed countries, while IMMIGRANT FROM LDC - the …rst and second generation immigrants from less developed countries. I include a set of variables showing the UI fund membership: UI FUND CONSTRUCTION, UI FUND MANUFACTURING, UI FUND TECHNICIANS, UI FUND TRADE, UI FUND CLERICAL, UI FUND ACADEMICS, OTHER UI FUND, and UI FUND SELF-EMPLOYED. Some of the UI funds exist based on the industry, while others are based on the educational achievements of the members. For example, UI FUND MANUFACTURING mainly insures unskilled workers of the manufacturing industry, and UI FUND ACADEMICS covers UI funds which insure academically educated workers. Thus, the UI membership to some extent repre- sents educational attainment, and to some extent - past occupation. I also had access to information on past labour market history. This includes data about the fraction of time spent on public income transfers for each of the past …ve years, and the

151 number of spells receiving public income transfers the person has experienced within the past …ve years. But only the information from the two last years, INCOME TRANSFER DEGREE LAST YEAR, INCOME TRANSFER DEGREE TWO YEARS AGO, # INCOME TRANS- FER SPELLS LAST YEAR, and # INCOME TRANSFER SPELLS LAST 2 YEARS, was signi…cantly di¤erent from zero in preliminary explorations, and therefore only information from these past two years before the unemployment spell is used. Finally, I also have access to information on sickness periods. The data from the last two years has been used in the analysis for the same reason as given above.

5. Econometric Model

The analysis is done using the timing-of-events model for identifying treatment e¤ects in a duration model framework developed by Abbring & van den Berg (2003). The timing-of-events model simultaneously models the transition rate out of unemployment and the transition rate into the ALMPs. The model is intended to correct for non-random selection into programmes with respect to observed as well as unobserved variables. Abbring & van den Berg (2003) show that with an assumption of 1) mixed proportional hazards and 2) a non-defective distribution of time until participation in ALMPs, given observed explanatory variables, the parameters of interest - say, the e¤ect of participation in ALMPs - are identi…ed non-parametrically. The implication is that there is no need for an exclusion restriction that is, a variable which appears in the selection equation, but which does not a¤ect the outcome variable, in this case the hazard rate out of unemployment. The intuition is that random variation in the timing of the event of participation in ALMPs separates the treatment e¤ect from the distribution of unobserved heterogeneity, which is assumed to be time-invariant.

Let Tu be a random variable denoting the duration of an unemployment spell, and let Tp be another random variable denoting the time from entry into unemployment until participation in the …rst ALMP. If we have Tp < Tu; a person participates in an ALMP during the unemployment spell. If Tp Tu, then Tp is censored, and the individual did not participate in an ALMP before  Tu.

Let X(t) be a vector of observed exogenous explanatory variables, and let Vu, and Vp =

152 (Vp1;Vp2;Vp3;Vp4) denote the unobserved variables possibly a¤ecting the exit rate out of unem- ployment and the entry rates into the four di¤erent types of ALMPs. The hazard into ALMPs is the sum of four cause-speci…c hazard rates, one for each type of ALMP: 4 p(tp x(tp); vp) = pi(tp x(tp); vpi): (1) j j i=1 X Each of these cause-speci…c hazards is assumed to be of the mixed proportional hazard type,

pi(tp x(tp); vpi) = pi(tp) exp x(tp) + vpi : (2) j pi 

Next, I de…ne two time varying vectors of indicator variables, d1(t) and d2(t): d1(t) is a 4 1  vector, where the i th element takes the value 1 if the individual participates in an ALMP of 0 type i at time t and takes the value 0 otherwise. Note that at most one element of d1(t) can take the value 1 at time t. Similarly, the i th element of d2(t) (which is also 4 1) takes the value 1 0  if the individual has completed an ALMP of type i during the last 26 weeks (the implication is that I only allow ALMPs to a¤ect the hazard rate out of unemployment up to 26 weeks after completion). Assuming once again a mixed proportional hazard rate, the hazard rate out of unemployment in the model is speci…ed as

u(tu x(tu); d1(tu); d2(tu); vu) j

= u(tu) exp [x(tu) u + d1(tu)1 + d2(tu)2 + vu] : (3)

The parameter 1 here measures the locking-in e¤ect, while 2 measures the post-programme e¤ect. In the estimations performed below, I will allow for separate e¤ects of programmes that start in di¤erent intervals of the UI bene…t spell, while for education and other ALMPs I run separate models to estimate the e¤ects, depending on the length of the programmes. But for expositional convenience, this interaction - between participation and completion indicators on the one side and the dummies of programme entry and programme length intervals on the other - has been ignored.

The timing-of-events model takes into account potential endogeneity of d1(t) and d2(t) by

153 allowing for correlation between the two unobserved components, Vu and Vp. That is, this method allows for selection on unobservables as well as observed explanatory variables.

I de…ne Cu as an indicator variable that takes the value 1 when the unemployment spells are completed and 0 for right censored unemployment spells, and so the contribution to the likelihood function of an individual with J unemployment spells, given observed and unobserved characteristics, is

J 1 tpj

exp p(s x(s); vp)ds u(t x(t); d1(t); d2(t); vu)dt ; (4)  2 j j 3 Z0 Z0 4 5 and the likelihood function can then be expressed as:

= (vu; vp)dG(vu; vp); (5) L L ZZ where G( ; ) is a bivariate distribution function for (vu; vp).   The expected duration of an unemployment spell may be calculated as

1 E[Tu x; d ; d ; v ] = S(t x; d ; d ; v )dt; (6) j 1 2 u j 1 2 u Z0 where the time-variation in the explanatory variables has been ignored for analytical conve- nience.

5.1. Identi…cation

Identi…cation in the timing-of-events model is based on several assumptions; one of these is a ’distributional’assumption requiring the hazard rates to be speci…ed as mixed proportional hazards. A ’noanticipation’assumption implies that the individual is allowed to know the distribution of time until programme participation and the distribution of programme types, but not the exact moment at which he will participate. If a person anticipates participation in an ALMP

154 at a particular future date tp he may use this information to decide about his current behaviour (for example he may decide to wait for the training by reducing his search intensity for jobs) and this may decrease the probability that Tu is quickly realized. If this is ignored in the empirical analysis the training e¤ect may be over-estimated (the opposite situation is also possible).

To identify the model there must be random variation in Tp at the individual level (this variation a¤ects Tu only by way of the treatment.). If persons enter a programme at, say, exactly one year after ‡owing into unemployment, it is impossible to distinguish the e¤ect of a programme from the duration dependence in the exit rate to work after one year. In such a case it is also hard to justify that entry into an ALMP is not anticipated.

The correlation between Vu and Vp is unrestricted in the model, which is important, since this is the correlation, which is intended to correct for selection. It is obvious that unob- served heterogeneity of the unemployed individuals plays an important role in the assignment to ALMPs. The variables taken into account by the caseworker (like motivation, subjectively assessed expected unemployment duration and subjective assessments of other aspects of the future career) are also indicative of unobserved determinants of the individual exit rate to work. Identi…cation of the timing-of-events model does not require exclusion restrictions8.

Endogeneity of the length of ALMPs is an important issue in model identi…cation. A number of existing studies tackle this question. For example, Hirano & Imbens (2004) employ a weak version of unconfoundedness assumption, i.e. conditional on observed covariates, the level of the treatment received (Ti) is independent of the potential outcome Yi(t). Under the weak unconfoundedness assumption, the average “dose response” function DRF (which is de…ned as the average e¤ect of the multi-valued or continuous treatment on the outcome of interest) could be derived by estimating average outcomes in subpopulations de…ned by pre-treatment covariates and di¤erent levels of the treatment. But then the dimensionality problem arises, and to solve it they introduce the generalized propensity score (GPS)9. The GPS method is recently used by Flores-Lagunes et al. (2007) to estimate the e¤ects

8 We should note that the variables that are observed by us and that may have an e¤ect on assignment to program are also observable to the individuals under consideration, so that we cannot impose exclusion restrictions, and we take the same vector x to a¤ect both u and p. 9 The GPS is de…ned as the conditional density of receiving a particular level of the treatment, t = T , which has the “balancing property”; i.e. within strata de…ned by values of GPS, the probability that t = T does not depend on the value of X.

155 of length of exposure to the Job Corps (JC) training programme in the US labour market. The study measures average causal e¤ects of di¤erent lengths of exposure to JC using the GPS under the assumption that the length of the individual’s JC spell is randomly assigned and …nds the estimated (marginal) e¤ects of an additional week of training to decline with the total length of enrolment in the programme. Cattaneo (2007) argues that in practice, however, treatments are frequently multi-valued and available econometric techniques in this literature cannot be applied directly. He studies the e¢ cient estimation of a large class of multi-valued treatment e¤ects as implicitly de…ned by a collection of possibly over-identi…ed non-smooth moment conditions when treatment assignment is assumed to be ignorable. The paper proposes two estimators; one based on an inverse probability weighting scheme and the other based on the e¢ cient in‡uence function of the model, and provides a set of su¢ cient conditions that ensure root-N consistency, asymptotic normality and e¢ ciency of these estimators. Among the studies focusing on duration analysis, Richardson & van den Berg (2008) consider a model in which the treatment e¤ect  is allowed to depend on the length of the program (t tp), the observed covariates (x) and the unobserved characteristics V:

(t tp; x; V) = (t tp) + x + V: 0 

Here V is allowed to be stochastically related to Vu and Vp. Since the exit rate to work is proportional to exp(), by analogy to the Mixed Proportional Hazard, they call the model “the Mixed Proportional Treatment E¤ect model”.

Modelling endogeneity in the model, however, would make the model very complicated; thus, this study assumes the length of programmes to be exogenous, given all included characteristics.

5.2. Parameterization

I assume all baseline hazard rates to be piecewise constant (that is j(t) = exp( jm), m =

1; :::; Mj , where Mj is the number of intervals for baseline hazard j). The following cut-o¤ points for the intervals are used for all hazard rates (the unemployment duration and time until programme participation are both measured in weeks): 4, 13, 26, 39, 52, 65, 78, 91, 104, 156.

156 With such a parametrization, it is straightforward to show that the expected duration of an unemployment spell is equal to

M u 1 E[Tu x; d ; d ; v ] = j 1 2 u hm(x; d ; d ; v ) m=1 i 1 2 u X P ( m 1 < Tu  m x; d1; d2; vu) ; (7)   j

m where the  m denote the cut-o¤ for the intervals, and hi the value of the hazard rate in interval m, and P ( m 1 < Tu  m x; d1; d2; vu) the probability that an individual leaves unemploy-  j ment in the m’th interval. For the mixture distribution, I apply a discrete distribution with two points of support 1 2 1 2 for each of the marginal distributions of the unobserved variables. Let (vu; vu) and (vpi; vpi), i = 1; 2; 3; 4 be the mass-points of Vu and Vpi, respectively. The associated probabilities are then: 1 1 1 1 1 P1 = P r(Vu = vu;Vp1 = vp1;Vp2 = vp2;Vp3 = vp3;Vp4 = vp4); 2 1 1 1 1 P2 = P r(Vu = vu;Vp1 = vp1;Vp2 = vp2;Vp3 = vp3;Vp4 = vp4); 1 2 2 2 2 P3 = P r(Vu = vu;Vp1 = vp1;Vp2 = vp2;Vp3 = vp3;Vp4 = vp4); 2 2 2 2 2 P4 = P r(Vu = vu;Vp1 = vp1;Vp2 = vp2;Vp3 = vp3;Vp4 = vp4); 4 with 0 Pi 1 for i = 1; ::::; 4, and Pi = 1. Note that the unobserved heterogeneity   i=1 terms are restricted to be perfectly correlatedX in the four cause-speci…c hazard rates into pro- grammes. This is also called a factor-loading speci…cation. It restricts the correlation between

Vpi and Vpj to be either 1 or 1 if i = j. The correlation between Vu and Vp is unrestricted, 6 which is important since this is the correlation which is intended to correct for selection on un- 1 observables. I normalize the distribution of the unobservables by setting vj = 0 for all hazard rates. This is done instead of normalising e.g. the mean of the mixture distribution to one.

6. Estimation Results

In this section, I cover the results of the analysis. The timing-of events model is estimated separately for the di¤erent age groups of the individuals (age 30-39, age 40-49 and age 50-59). Separate models are also estimated to examine the e¤ects of the ALMPs, depending on the time spent in unemployment before entry into the programme, and to discover the e¤ects of

157 Education and Other ALMPs, depending on the length of the programme. While modelling the dependence of the impact of ALMPs on programme entry time, I specify the following intervals of the unemployment spell: 1-6 months, 7-12 months, 13-18 months, 19-24 months and >24 months in unemployment. To measure the length-dependent e¤ects, I distinguish the following intervals of program length: 1-6 weeks, 7-12 weeks, 13-18 weeks, 19-24 weeks and >24 weeks. Finally I estimate the e¤ects of Education and Other ALMPs lasting 6 weeks or shorter (which are found to reduce the length of the UI bene…t spell), depending on the time spent in unemployment before entry into the programme.

Table 2. Modelling assumptions

ALMPs e¤ects depending on introduction time Interval of UI Activation is assumed Progr. length assumed 1-6 months After 13 weeks on UI 26 weeks (Private 7-12 months After 39 weeks on UI and Public Job Training); 13-18 months After 52 weeks on UI 16 weeks (Education); 19-24 months After 78 weeks on UI 8 weeks (other ALMPs) >24 months After 104 weeks on UI ALMPs e¤ects depending on programme length Interval of UI Activation is assumed Progr. length assumed 1-6 weeks 4 weeks 7-12 weeks 8 weeks 13-18 weeks After 52 weeks on UI 14 weeks 19-24 weeks 20 weeks >24 weeks 26 weeks

The e¤ects have been averaged using the methodology explained in section 5. ’Locking- in e¤ect’and ’Post-program e¤ect’in the tables show the coe¢ cients estimated; that is, the exponential function of the reported parameter gives the multiplicative impact on the hazard rates. The ’Nete¤ect’shows the e¤ect of the programmes on expected unemployment duration. The net e¤ect is calculated for a ’standard person’with the average characteristics for each of the samples used, and I employ a set of assumptions shown in Table 2. Unfortunately, I cannot measure the e¤ectiveness of the ALMPs depending on the planned length of the programme. Only the data about the programmes that actually took place is available. Since there is no information on planned course lengths, there is also no information

158 about the completion of the course. It is possibly the case that those who end the course soon after entry do this because they …nd a job and this creates a spurious correlation between length of ALMP and the hazard rate. However, if the actual length of the programme was shorter than the planned one then we should expect a positive bias in the locking-in e¤ect. We would not expect a large e¤ect on the post-programme e¤ect parameter.

In the sub-section 6.1. below, I present and discuss the e¤ects of the ALMPs depending on the time of programme entry. Sub-section 6.2. covers the results dependent on the length of Education and Other ALMPs, while the e¤ects of Education and Other ALMPs lasting 6 weeks or shorter are discussed in sub-section 6.3. Below I graph the net e¤ects of the programmes; that is, I show how participation in the programmes reduced or increased unemployment duration. To see the estimation results of locking-in, post-programme and net e¤ects of the programmes, I refer to tables A.1. - A.8. in Appendix.

6.1. ALMPs E¤ects Depending on Entry Time

In this sub-section, I present and discuss the estimation results of the e¤ectiveness of ALMPs depending on time of programme entry. In Figures 1-4, I graph the net e¤ects of the programmes on unemployment duration, while the values of the estimated coe…cients of locking-in, post- program and net e¤ects are presented in Tables A.1-A.4 in Appendix. Concerning the e¤ectiveness of Private job training (Figure 1 and Table A.1.), I …nd these programmes to increase unemployment duration of the 30-39 and 40-49 age group representa- tives (by 0.73 and 1.19 weeks, respectively) when the programme is assigned in the …rst half year of unemployment. However, programme, introduced in the interval of 7-12 months of the UI bene…t spell gives a slightly positive response for the 30-39 age group (the net e¤ect of -0.37 weeks), while the biggest e¤ect from Private job training is found when applying it after 1 year spent unemployed. Then the programme reduces unemployment duration by 1.18 weeks for 30-39-year-old persons and by 1 week for 40-49-year-old. When introduced in 19-24 months of UI bene…t spell or later, the programme still reduces the unemployment duration; but its e¤ectiveness decreases. Private Job training is much more e¤ective for elderly individuals (the 50-59 age group). Even in case of very early assignment (in the …rst 6 months of unemployment), the programme

159 reduces unemployment duration by 2.69 weeks, while the biggest e¤ect (the net e¤ects of 5.35 - 6.4 weeks) is found if it is introduced in the 7-24 months interval of an unemployment spell. The estimated e¤ects of Public job training (Figure 2) are not nearly as positive. Though moderate positive post-programme e¤ects were estimated in some cases, they are not nearly su¢ cient to compensate the dramatic locking-in e¤ects (Table A.2.). Thus, the programme increases unemployment duration for all age groups of the individuals, and is found to be the most harmful for the elderly unemployed population. Public job training shows the worst performance when assigned in the …rst 6 months of an UI bene…t spell. The negative e¤ects of the programme gradually decrease (but remain negative) when the activation takes place at later stages of unemployment. Education (Figure 3 and Table A.3.) and Other ALMPs (Figure 4 and Table A.4.) perform worst in the very early stage of unemployment. When education takes place in the …rst half year of an UI bene…t spell, it increases unemployment duration by 4.14 weeks, 4.93 weeks and 9.43 weeks for the 30-39, 40-49 and 50-59 age groups, respectively. The corresponding …gures for Other ALMPs are 1.86 weeks, 2.47 weeks and 5.48 weeks. The performance of these two groups of ALMPs improves when the activation takes place after 6 months spent unemployed. The net e¤ect of the programmes is then slightly positive (the programmes increase unemployment duration), while for the 50-59 age group the other ALMPs reduce unemployment duration by 1.19 weeks. Other ALMPs, assigned in 13-18 months of UI bene…t spell, decrease unemployment dura- tion, but very slightly, while the e¤ect of education is an increase in unemployment (the net e¤ects of 0.9 weeks (30-39 group), 1.02 weeks (40-49 group) and 1.82 weeks (50-59 group)). And …nally, both programmes become ine¤ective when the activation takes place after 2 years spent on UI bene…ts.

160 Figure 1. Net e¤ect of Private job training (based on programme entry time, months)

2,00 1•6 1,00 7•12 13•18 19•24 >24 0,00 •1,00 •2,00

•3,00 Weeks •4,00 •5,00 •6,00 •7,00

30•39 40•49 50•59

Figure 2. Net e¤ect of Public job training (based on programme entry time, months)

9,00 8,00

7,00 6,00 5,00

4,00 Weeks 3,00

2,00

1,00 0,00 1•6 13•18 7•12 19•24 >24 30•39 40•49 50•59

161 Figure 3. Net e¤ect of Education (based on programme entry time, months)

10,00

8,00

6,00

Weeks 4,00

2,00

0,00 1•6 7•12 13•18 19•24 >24 30•39 40•49 50•59

Figure 4. Net e¤ect of Other ALMPs (based on programme entry time, months)

6,00

4,00

1•6 2,00 Weeks

0,00 7•12 13•18 19•24 >24

•2,00

30•39 40•49 50•59

162 The results above seem to favour activation of unemployed in their …rst year on UI bene…t. However, a very early activation is not supported. Neither of the programmes showed favourable performance in the …rst half year of unemployment (with the exception of Private Job training to 50-59 old persons). Having in mind the increasing costs of the programmes, since the target group of participants becomes wider (30-35% of all activations done during the period under study belonged to activations done in 1-6 months of UI bene…t spell), we can conclude that the activation of unemployed persons in their …rst half year of UI bene…t spell does not lead to the e¢ cient allocation of the means intended to the Active Labour Market Measures in Denmark. Only in one of the four groups of Danish ALMPs - Private job training - is the post- programme e¤ect large enough to counteract the locking-in e¤ect, and in most cases the pro- gramme reduces unemployment duration, while the participants in the other three types of programmes experience increasing unemployment duration (with some exceptions in case of other ALMPs). This is in line with …ndings in the previous studies on Danish data (see Bolvig et al. (2003), Graversen (2004), Rosholm & Lauzadyte (2008)). In addition, private job training programmes are found to be highly e¤ective for the 50-59 old individuals, who are the weakest group of the unemployed population. Thus, a more active use of this programme to the elderly workers should be promoted. And …nally, it is disappointing that after two years spent unemployed the ALMPs become ine¤ective (with the exception of Private job training for 50-59-year-old individuals), considering that under the current Active Labour Market Policies in Denmark every person, who is still unemployed after two and a half years, has the obligation of full time participation in ALMPs during the rest time of UI bene…t spell.

6.2. ALMP E¤ects Depending on Programme Length

A study on the e¤ectiveness of Education and Other ALMPs depending on the length of the programmes (Figures 5 and 6 and tables A.5 - A.6.) leads to the results that only programmes lasting 6 weeks or shorter reduce unemployment durations of the individuals belonging to all age groups. The oldest age group experienced the strongest positive e¤ects of the programmes - their unemployment duration was reduced by 4.3 weeks and 6.7 weeks by Education and Other ALMPs respectively. For younger persons the programmes are found to be bene…cial as well;

163 with a net e¤ect of about 1 week in case of Education and about 2 weeks in case of Other ALMPs. The favourable e¤ects of the programmes are the result of large post-programme e¤ects, which were able to compensate the moderate negative locking-in e¤ects. That is, the hazards into employment increased strongly after programme completion. Other ALMPs of 7-12 weeks length are found to produce zero net e¤ect for 40-49-year-old persons (the positive post-programme e¤ect just covered the negative locking-in e¤ect, but was not large enough to in‡uence the unemployment duration). For the individuals in 30-39 and 50-59 age groups, the programme reduced the length of the UI spell, though very slightly. However, education programs longer than 6 weeks and Other ALMPs longer than 12 weeks were found to increase unemployment duration for all age groups. It is disappointing that a negative relationship was found between the length and the e¤ectiveness of the programmes: the longer programme duration, the stronger the negative e¤ect produced. Further, the programmes do not produce any positive e¤ects in the period after activation - the post programme e¤ects for education of more than 6 weeks and other ALMPs of 13 weeks or longer are found to be negative; and sometimes they are close to the locking-in e¤ect. The negative performance of the programmes does not vary between the younger age groups, while the harm to the elderly UI bene…ts recipients is much stronger. There could be several reasons for such programme e¤ects. Firstly, the information about the planned length of ALMPs is not available, so it could be the case that the actual length of the programme was shorter than the planned one if the programmes were cancelled by the individuals having found a job. However, then we should expect a positive locking-in e¤ect, which is not the case in our situation. Thus, this does not seem to be an appropriate explanation of the favourable behaviour of short-term education and Other ALMPs in Denmark.

The motivation to become employed and the personal characteristics of the persons who enter ALMPs may di¤er as well. In this model, however, the length of the programmes is assumed to be exogenous (see sub-section 5.1. on model identi…cation). If in reality it is the case that individuals with better personal characteristics enter the short term programmes, the selectivity issue arises and the e¤ects of these programmes may be overestimated.

164 Figure 5. Net e¤ect of Education (based on programme length, weeks)

8,00

6,00

4,00

2,00 1•6 0,00 Weeks 7•12 13•18 19•24 >24 •2,00

•4,00

•6,00

30•39 40•49 50•59

Figure 6. Net e¤ect of Other ALMPs (based on programme length, weeks)

6,00

4,00

2,00 1•6 7•12 0,00 13•18 19•24 >24 •2,00 Weeks

•4,00

•6,00

•8,00

30•39 40•49 50•59

165 This issue becomes event more important in light of the fact that under the existing ac- tive labour market policies in Denmark an unemployed person has the right to a choice of a programme of up to 6 weeks duration during the passive period of his UI bene…t spell. To have a complete picture of the situation, take a look to the popularity of di¤erent lengths of programmes among the activated individuals. Table 3 shows that about one third of those who had training, were in courses of up to six weeks duration (for persons over …fty the …gure come close to 40%). The share of medium length courses’varies among age groups, while close to one fourth of the trainees had more than half a year lasting training (which was the most popular among 30-39 years old individuals). Concerning education programmes, we can see that the distribution of length of the pro- grammes is rather balanced. This is not the case for Other ALMPs, where about 70% of activations are short activations (and about 20% of individuals were activated in 7-12 weeks lasting programmes). The shares of those who took part in Other ALMPs of 13 weeks and longer are very low.

Table 3. Distribution of Length of Education and Other ALMPs,%*

Programme Education Other ALMPs length 30-39 40-49 50-59 Total 30-39 40-49 50-59 Total 1-6 weeks 31.2 35.5 38.3 34.4 75.8 69.8 65.1 70.7 7-12 weeks 12.9 14.5 19.2 15.1 16.8 22.8 19.2 19.4 13-18 weeks 12.4 18.3 10.9 13.8 3.3 3.0 6.0 4.0 19-24 weeks 16.1 10.2 10.4 12.7 3.2 2.7 4.3 3.3 >24 weeks 27.4 21.5 21.2 23.9 0.9 1.8 5.4 2.5

* Individuals activated in a particular type of ALMPs sum up to 100%.

To investigate this issue further, I estimate the e¤ectiveness of the short-term Education and Other ALMPs, depending on programme entry time. These estimations partially support the idea above - the programmes of up to 6 weeks are found to perform best when the activation takes place in the second half year of unemployment; while Other ALMPs are found to be extremely e¤ective for persons over …fty in the very early stage of activation. On the other hand, the short-term programmes behaved well in the second year of the UI bene…ts spell as well, while for the elderly individuals this was the case also after two years of unemployment

166 (for detailed results, see sub-section 6.3). Thus, the e¤ectiveness of these programs should not be explained by the motivation of the participants alone.

6.3. E¤ects of ALMPs of 1-6 Weeks Length Depending on Entry Time

Turning to the e¤ectiveness of education of 1-6 weeks length for 30-39-year-old individuals, I …nd the programme to be the most e¤ective when it is assigned in the second half year of unemployment (the programme reduces unemployment duration by 2.81 weeks). But the short courses are also found to be e¤ective in 13-18 months of unemployment (1.97 weeks UI spell reduction). In the …rst half year of UI bene…t spell and after 2 years spent on UI bene…ts, the courses do not have an impact, while in 19-24 months of UI spell they produce a moderate positive behaviour (the net e¤ect of -0.53 weeks). Other ALMPs also are found to be the most e¤ective when activation of 30-39-year-old persons is done in 7-12 months and in 13-18 months of UI bene…t spell (the net e¤ects are -3.33 and 2.04 respectively). However, the programmes perform quite well in 1-6 months as well as in 19-24 of unemployment. After 2 years on UI bene…ts, neither short courses nor Other ALMPs of 1-6 weeks length give any positive results. Analysis of the e¤ects of ALMPs, lasting 6 weeks or shorter, for 40-49 age group reveals that short training is e¤ective when it is assigned in the …rst year of unemployment, but show the best results after a half year spent on UI bene…ts (unemployment duration is reduced by almost three weeks). The programme produces a moderate positive performance in 13-38 months of UI bene…t spell (the net e¤ect of -0.68), while after 1.5 year of UI bene…t receipt becomes ine¤ective. Performance of other ALMPs of 1-6 weeks length for 40-49-year-old persons is similar to the performance for 30-39-year-old ones. The most e¤ective programme is in 7-18 months of the UI bene…t spell, but also behaves quite well in 1-6 and 19-24 months of unemployment. Di¤erently from the age groups above, for 50-59-year-old individuals both education and other ALMPs are found to be the most e¤ective in the 1st year of UI bene…t spell. Already in the …rst half year on UI bene…ts, the courses reduce unemployment duration of these individuals by 4.37 weeks. The most e¤ective programme is found to be in 7-12 months of unemployment, when it produces the net e¤ect of -5.86 weeks.

167 Figure 7. Net e¤ect of 1-6 weeks Education (based on programme entry time, months)

2,00 1•6 7•12 13•18 19•24 >24 0,00

•2,00 Weeks •4,00

•6,00

30•39 40•49 50•59

Figure 8. Net e¤ect of 1-6 weeks Other ALMPs (based on programme entry time, months)

1•6 7•12 13•18 19•24 >24 0,00

•2,00

•4,00 Weeks

•6,00

•8,00

30•39 40•49 50•59

168 After 1 year of UI bene…t receipt, the e¤ectiveness of short length education decreases, but the programme remains positive in the later stages of unemployment (decreases the length of UI bene…t spell by more than 3 weeks). Other ALMPs of 6 weeks or shorter are found to perform very well in the …rst half year of UI bene…t spell - it reduces unemployment duration of 50-59-year-old persons by almost 8 weeks. The net e¤ect of the programme decreases (by almost 2 weeks) after a half year spent unemployed; however, the programme remains e¤ective in all stages of the UI bene…t spell, and even after 2 years spent on UI bene…ts, it reduces unemployment duration by almost 4 weeks.

The analysis of the e¤ects of ALMPs of 1-6 weeks length, depending on entry time, leads to the conclusion that both programmes - Education and Other ALMPs - are found to be the most e¤ective for elderly unemployed persons in all stages of the UI bene…t spell. However, the programmes produce a favourable net e¤ect for the younger individuals as well. Furthermore, a favourable net e¤ect of the programmes is found to be the result of a high post-programme e¤ect, which in most cases compensates a moderate negative locking-in e¤ect10. The results above encourage to foster activation of 50-59-year-old persons in the …rst year of unemployment. The second half year of the UI bene…t spell is found to be the best time to activate 30-49-year-old individuals, but a very early activation of them could also be supported (with the exception of short courses for the 30-39 age group). The estimations above discover that the e¤ectiveness of the short-term programmes should not be explained by the motivation of the participants alone, and encourage to have a deeper look into the types and intensity of these programmes in Denmark, i.e. what types of training are covered by the short courses, and which ALMPs of 1-6 weeks length were the most popular among the di¤erent age groups of the unemployed individuals, and in di¤erent stages of the UI bene…t spell. And …nally, like in the general case of entry-dependent ALMP e¤ects (see sub-section 6.1.), the ALMPs lasting 6 weeks or shorter become ine¤ective for 30-49-year-old individuals after two years of UI bene…ts receipt. Thus, the long-term unemployed in Denmark seem to have speci…c problems which cannot be addressed by participation in the programmes.

10 Only Other ALMPs of 1-6 weeks length, assigned to 50-59-year-old persons in 7-12 months of their UI bene…t spell, produce a small positive locking-in e¤ect.

169 7. Conclusions

In this study, I estimate the e¤ects of ALMPs on UI bene…t recipients in Denmark. Firstly, I examine the e¤ects of the programmes, depending on the time spent in unemployment before entry into the programme. Then I analyse how the performance of the two types of ALMPs - Education and Other ALMPs - varies with the length of the programmes, and lastly, I carry out the programme entry time dependent analysis of the short-term Education and Other ALMPs. I use the timing-of-events model introduced by Abbring and van den Berg (2003). This approach models the process of exit from unemployment into employment and the process of programme entry simultaneously in a multivariate hazard model, and I estimate two separate e¤ects of the ALMPs in a duration model: a locking-in e¤ect and a post-programme e¤ect. Finally, I calculate the net e¤ects of the ALMPs on unemployment duration. I …nd that in only one of the four groups of Danish ALMPs - Private Job training - is the post-program e¤ect large enough to counteract the locking-in e¤ect, and in most cases the programme reduces unemployment duration while the participants in the other three types of programmes in most cases experience increasing unemployment duration. The programme entry time dependent analysis leads to a conclusion in favour of activation of unemployed individuals in the second half year of the UI bene…t spell, but very early activation is not supported by the results, while the analysis of the impact of Education and Other ALMPs depending on the length of the programmes …nds that only the short-term programmes of up to 6 weeks reduce unemployment duration. Persons over …fty are found to be the most sensitive group to the in‡uence (both positive and negative) of ALMPs, and private job training programmes as well as the short-term activation are found to be highly e¤ective in all stages of their unemployment spell. However, for the 30-49-year-old individuals in Denmark all types of programs are found to be ine¤ective after two years on UI bene…ts. Thus, the long-term unemployed in Denmark seem to have speci…c problems which cannot be addressed by participation in programmes. These results motivate for further explorations: …rstly, which types of programmes have been e¤ective for the particular groups of individuals at di¤erent stages of their UI bene…t spell, and secondly, the ’threats’of the programmes should be considered.

170 References

[1] Abbring, J. and G. van den Berg (2003). “The Non-Parametric Identi…cation of Treatment E¤ects in Duration Models”, Econometrica, 71, 1491-1517.

[2] Black, D., J. A. Smith, M. C. Berger and B. J. Noel (2003). "Is the Threat of Reem- ployment Services More E¤ective than the Services Themselves?", American Economic Review, Vol. 93, 1313-1327.

[3] Blundell, R., M. Costa Dias, C. Meghir and J. Van Reenen (2004). Evaluating the employment impact of a mandatory job search program, Journal of the European Economic Association, 2, 569-606.

[4] Bolvig, I., P. Jensen and M. Rosholm (2003). "The Employment E¤ects of Active Social Policy", IZA Discussion Paper No. 736, Institute for the Study of Labor, Bonn.

[5] Brodaty, T., B. Crepon and D. Fougere (2002). "Do Long-Term Unemployed Workers Bene…t from Active Labor Market Programs? Evidence from France, 1986-1998", mimeo.

[6] Carling, K and L. Larsson (2005). “Does early intervention help the unemployed youth?”, Labour Economics, vol 12, nr 3, 301-319.

[7] Cattaneo, M.D. (2007). "E¢ cient Semiparametric Estimation of Multi-valued Treatment E¤ects". Manuscript, Univ. of Michigan.

[8] Fay, R. G. (1996). "Enhancing the E¤ectiveness of Active Labour Market Policies: Evidence from Programme Evaluations in OECD Countries". OECD Labour Market and Social Policy Occasional Papers, No. 18, OECD Publishing.

[9] Flores-Lagunes, A., A. Gonzalez and Todd C. Neumann (2007). "Estimating the E¤ects of Length of Exposure to a Training Program: The Case of Job Corps," IZA Discussion Papers 2846, Institute for the Study of Labor (IZA)

[10] Graversen, B. (2004). Employment e¤ects of active labor market programs: Do the programs help welfare bene…t recipients to …nd jobs?, PhD thesis 2004-2, Department of Eco- nomics, University of Aarhus.

171 [11] Heckman, J.J., R.J. Lalonde and J.A. Smith (1999). The economics and econometrics of active labor market programs, in: Ashenfelter, O. and D. Card (eds.) Handbook of Labor Economics, Volume 3A, Amsterdam, North-Holland.

[12] Hirano, K. and G.W. Imbens (2004). "The Propensity Score with Continuous Treat- ments." In Andrew Gelman and Xiao-Li Meng (eds.), Applied Bayesian Modeling and Causal Inference from Incomplete-Data Perspectives. West Sussex: John Wiley and Sons, 73-84.

[13] Jensen, P. and Westergård-Nielsen, N., 1990. Temporary layo¤s. In: Hartog, J., Ridder, G. and Theeuwes, J., Editors, 1990. Panel Data and Labor Market Studies, Elsevier, pp. 73– 189.

[14] Lauzadyte, A. and M. Rosholm (2006). "A Meta Analysis of County, Gender, and Year Speci…c E¤ects of Active Labour Market Programmes", Working Paper, University of Aarhus.

[15] Mealli, F., S. Pudney and J.Thomas (1996). "Training Duration and Post-training Outcomes: A Duration-Limited Competing Risks Model," Economic Journal, Royal Economic Society, vol. 106(435), pages 422-33, March

[16] Munch, J. and L. Skipper (2004). "The consequences of active labor market program participation in Denmark", Working Paper, University of Aarhus.

[17] Richardson, K. and G.J. van den Berg (2008). "Duration dependence versus unobserved heterogeneity in treatment e¤ects: Swedish labor market training and the transition rate to employment". IFAU working paper 2008:7, p. 39-40.

[18] Rosholm, M. and M. Svarer (2004). “Estimating the Threat E¤ect of Active Labour Market Programmes”, Manuscript. University of Aarhus.

[19] Rosholm, M. (2007). "Experimental Evidence of the Nature of the Danish Employment Miracle". Manuscript. University of Aarhus.

[20] Torp, H. (1994). "The Impact of Training on Employment: Assessing a Norwegian Labour Market Programme". Scandinavian Journal of Economics, vol. 96, no. 4, pp. 531-550.

172 [21] Upward, R. (2002). "Evaluating outcomes from the Youth Training Scheme using matched …rm-trainee data". Oxford Bulletin of Economics and Statistics, Vol. 64, No 3, July 2002 , pages 277-306 (30).

[22] Van den Berg, G. J. (2001). "Duration models: speci…cation, identi…cation, and multiple durations in: Heckman, J.J. and E. Leamer (eds) Handbook of Econometrics, Volume V, North- Holland.

[23] Weber, A. and H. Hofer (2004(a)). "Are Job Search Programs a Promising Tool? A Microeconometric Evaluation for Austria". IZA Discussion Paper No 1075.

[24] Weber, A. and H. Hofer (2004(b)) "Employment E¤ects of Early Interventions on Job Search Programs" IZA Discussion Paper No 1076.

173 APPENDIX

TABLE A.1. EFFECTS OF PRIVATE JOB TRAINING (PROGRAMME INTRO TIME)

Months in Locking•in effect Post•programme effect Net effect (weeks) unemployment Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. 1•6 30•39 •0.42 0.10 •34.4 0.41 0.07 50.1 0.73 0.15 40•49 •0.40 0.14 •33.1 0.33 0.09 39.0 1.19 0.36 50•59 •0.14 0.02 •13.4 0.50 0.09 65.4 •2.69 0.47 7•12 30•39 •0.20 0.03 •18.1 0.46 0.16 58.6 •0.37 0.07 40•49 •0.14 0.23 •13.2 0.09 0.21 9.2 0.23 0.44 50•59 0.21 0.02 23.7 0.51 0.22 66.0 •5.27 0.96 13•18 30•39 0.09 0.02 9.7 0.74 0.17 110.3 •1.18 0.28 40•49 0.26 0.29 29.8 0.44 0.23 54.9 •1.00 0.71 50•59 0.42 0.31 52.3 0.84 0.24 131.4 •6.12 1.01 19•24 30•39 0.12 0.06 12.7 0.86 0.20 137.1 •0.35 0.11 40•49 •0.06 0.03 •5.4 0.76 0.25 114.6 •0.05 0.02 50•59 0.50 0.31 64.7 0.71 0.36 102.8 •6.37 1.25 >24 30•39 •0.11 0.03 •10.6 0.73 0.18 108.0 •0.34 0.07 40•49 •0.12 0.03 •11.0 0.63 0.21 87.0 •0.24 0.08 50•59 0.34 0.14 40.0 0.61 0.25 83.8 •3.21 0.84

TABLE A.2. EFFECTS OF PUBLIC JOB TRAINING (PROGRAMME INTRO TIME)

Months in Locking•in effect Post•programme effect Net effect (weeks) unemployment Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. 1•6 30•39 •0.78 0.10 •54.2 0.20 0.06 22.0 3.00 0.55 40•49 •1.12 0.11 •67.3 0.30 0.07 34.8 4.22 0.56 50•59 •1.34 0.13 •73.8 0.10 0.08 10.1 8.21 1.43 7•12 30•39 •0.62 0.17 •46.3 0.10 0.02 10.8 1.21 0.25 40•49 •0.69 0.16 •49.8 •0.09 0.02 •8.6 1.69 0.36 50•59 •0.86 0.18 •57.8 •0.15 0.02 •14.1 5.78 0.97 13•18 30•39 •0.75 0.21 •53.0 •0.36 0.19 •30.5 1.80 0.65 40•49 •0.88 0.20 •58.6 0.14 0.15 15.0 1.14 0.43 50•59 •0.81 0.21 •55.3 0.04 0.02 3.7 3.14 1.07 19•24 30•39 •0.34 0.22 •29.0 •0.05 0.02 •4.9 0.39 0.19 40•49 •0.84 0.24 •56.8 •0.06 0.02 •5.6 0.72 0.22 50•59 •0.76 0.27 •53.1 0.14 0.03 14.8 1.66 0.42 >24 30•39 •0.44 0.18 •35.4 0.15 0.04 16.5 0.18 0.05 40•49 •0.65 0.22 •47.8 0.06 0.01 5.8 0.30 0.09 50•59 •0.50 0.24 •39.5 •0.08 0.18 •7.9 1.11 0.87

174 TABLE A.3. EFFECTS OF EDUCATION (PROGRAMME INTRO TIME)

Months in Locking•in effect Post•programme effect Net effect (weeks) unemployment Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. 1•6 30•39 •1.16 0.05 •68.5 •0.16 0.03 •15.0 4.14 0.27 40•49 •1.19 0.06 •69.4 •0.19 0.04 •17.6 4.93 0.38 50•59 •1.10 0.07 •66.7 •0.25 0.05 •21.9 9.43 0.91 7•12 30•39 •0.84 0.14 •56.8 0.05 0.01 5.3 0.42 0.07 40•49 •0.59 0.16 •44.7 •0.12 0.11 •11.5 0.43 0.18 50•59 •0.64 0.25 •47.1 0.13 0.01 13.4 0.73 0.12 13•18 30•39 •0.85 0.10 •57.2 0.02 0.07 2.1 0.90 0.21 40•49 •0.80 0.13 •55.3 •0.12 0.09 •11.2 1.02 0.26 50•59 •0.62 0.17 •46.1 •0.01 0.01 •1.2 1.82 0.76 19•24 30•39 •0.98 0.09 •62.4 0.10 0.06 10.8 0.98 0.15 40•49 •0.81 0.10 •55.4 0.01 0.07 1.4 1.07 0.26 50•59 •0.73 0.14 •51.9 •0.04 0.02 •3.9 3.24 0.92 >24 30•39 •0.83 0.14 •56.2 0.18 0.08 19.8 0.16 0.04 40•49 •1.00 0.21 •63.3 0.15 0.10 15.7 0.20 0.06 50•59 •1.10 0.35 •66.6 0.34 0.12 40.8 0.14 0.05

TABLE A.4. EFFECTS OF OTHER ALMPS (PROGRAMME INTRO TIME)

Months in Locking•in effect Post•programme effect Net effect (weeks) unemployment Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. 1•6 30•39 •0.50 0.07 •39.4 •0.19 0.04 •17.1 1.86 0.33 40•49 •0.46 0.09 •36.9 •0.29 0.05 •24.9 2.47 0.46 50•59 •0.66 0.11 •48.5 •0.18 0.06 •16.8 5.48 0.87 7•12 30•39 •0.12 0.02 •11.3 •0.06 0.11 •5.4 0.19 0.06 40•49 •0.43 0.25 •34.8 •0.01 0.01 •1.3 0.29 0.21 50•59 0.01 0.14 1.0 0.48 0.23 61.49 •1.19 1.10 13•18 30•39 •0.01 0.13 •0.6 0.13 0.02 14.4 •0.05 0.02 40•49 0.08 0.15 8.2 0.11 0.03 11.4 •0.11 0.05 50•59 •0.01 0.19 •0.8 0.13 0.03 14.3 •0.14 0.07 19•24 30•39 •0.27 0.16 •23.4 0.12 0.08 12.4 0.07 0.05 40•49 •0.14 0.17 •13.0 0.01 0.01 1.0 0.11 0.12 50•59 •0.22 0.02 •19.4 •0.06 0.01 •6.2 0.98 0.12 >24 30•39 •0.35 0.27 •29.8 0.16 0.13 16.9 0.03 0.02 40•49 •0.16 0.26 •15.1 0.06 0.15 6.6 0.00 0.00 50•59 •0.04 0.02 •4.3 0.09 0.03 9.2 •0.03 0.01

175 TABLE A.5. EFFECTS OF EDUCATION (BASED ON PROGRAMME DURATION)

Locking•in effect Post•programme effect Net effect (weeks) Program duration Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. 1•6 30•39 •0.36 0.09 •30.5 0.48 0.03 61.0 •0.86 0.10 40•49 •0.31 0.10 •26.7 0.58 0.04 78.9 •1.06 0.12 50•59 •0.26 0.12 •23.1 0.70 0.05 102.1 •4.29 0.55 7•12 30•39 •0.35 0.08 •29.5 •0.16 0.05 •14.5 0.49 0.14 40•49 •0.41 0.11 •33.9 •0.26 0.07 •22.7 0.65 0.17 50•59 •0.47 0.13 •37.4 •0.12 0.08 •11.3 1.28 0.51 13•18 30•39 •0.58 0.09 •43.8 •0.37 0.07 •31.0 1.12 0.19 40•49 •0.62 0.11 •46.1 •0.60 0.08 •44.9 1.36 0.21 50•59 •0.97 0.16 •62.1 •0.52 0.09 •40.5 4.15 0.70 19•24 30•39 •1.20 0.10 •69.8 •0.67 0.08 •49.0 2.26 0.23 40•49 •1.19 0.13 •69.6 •0.82 0.10 •56.1 2.28 0.26 50•59 •1.19 0.18 •69.6 •0.81 0.12 •55.4 6.13 0.94 >24 30•39 •1.39 0.05 •75.0 •0.93 0.06 •60.4 2.97 0.15 40•49 •1.35 0.07 •74.2 •1.17 0.08 •69.0 2.99 0.18 50•59 •1.16 0.09 •68.7 •1.08 0.10 •65.9 7.49 0.64

TABLE A.6. EFFECTS OF OTHER ALMPS (BASED ON PROGRAMME DURATION)

Locking•in effect Post•programme effect Net effect (weeks) Program duration Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. 1•6 30•39 •0.29 0.13 •25.4 0.71 0.05 102.9 •1.90 0.24 40•49 •0.15 0.14 •13.8 0.70 0.06 102.2 •1.79 0.30 50•59 •0.20 0.09 •17.7 0.91 0.07 149.1 •6.72 0.87 7•12 30•39 •0.18 0.12 •16.2 0.13 0.07 13.3 •0.13 0.08 40•49 •0.11 0.14 •10.3 0.08 0.50 8.6 0.01 0.03 50•59 •0.12 0.17 •11.2 0.06 0.10 6.2 •0.09 0.13 13•18 30•39 •0.14 0.13 •12.8 •0.12 0.08 •10.9 0.42 0.34 40•49 •0.32 0.11 •27.4 •0.19 0.16 •17.3 0.77 0.37 50•59 •0.46 0.21 •36.7 •0.17 0.12 •16.0 2.13 1.14 19•24 30•39 •0.45 0.18 •36.5 •0.42 0.09 •34.4 1.50 0.43 40•49 •0.65 0.12 •48.0 •0.50 0.23 •39.2 1.73 0.46 50•59 •0.88 0.31 •58.6 •0.60 0.15 •45.4 5.33 0.97 >24 30•39 •0.66 0.30 •48.1 •0.52 0.16 •40.6 2.16 0.80 40•49 •0.57 0.19 •43.4 •0.30 0.03 •25.9 1.36 0.21 50•59 •0.98 0.20 •62.4 •0.54 0.33 •41.9 5.67 1.04

176 TABLE A.7. EFFECTS OF 1-6 WEEKS EDUCATION (PROGRAMME INTRO TIME)

Months in Locking•in effect Post•programme effect Net effect (weeks) unemployment Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. 1•6 30•39 •0.28 0.08 •24.1 0.23 0.15 25.9 •0.04 0.02 40•49 •0.19 0.07 •16.9 0.52 0.14 67.7 •1.22 0.37 50•59 •0.25 0.09 •21.8 0.71 0.08 104.3 •4.37 0.78 7•12 30•39 •0.24 0.17 •21.3 0.90 0.08 145.7 •2.81 0.45 40•49 •0.10 0.02 •9.7 0.86 0.37 136.1 •2.91 0.81 50•59 •0.25 0.13 •22.3 0.91 0.43 149.5 •5.86 2.14 13•18 30•39 •0.27 0.18 •23.4 0.80 0.19 121.5 •1.97 0.69 40•49 •0.24 0.19 •21.2 0.40 0.11 49.1 •0.68 0.24 50•59 •0.16 0.10 •15.0 0.69 0.22 99.2 •3.83 1.56 19•24 30•39 •0.31 0.21 •26.8 0.45 0.15 56.9 •0.53 0.24 40•49 •0.27 0.15 •23.9 0.34 0.14 39.8 •0.15 0.08 50•59 •0.40 0.34 •33.2 0.69 0.30 99.0 •3.10 1.73 >24 30•39 •0.72 0.22 •51.4 0.39 0.20 48.2 0.03 0.01 40•49 •0.57 0.23 •43.7 0.60 0.11 81.4 •0.17 0.04 50•59 •0.32 0.36 •27.7 0.66 0.21 92.6 •3.32 1.64

TABLE A.8. EFFECTS OF 1-6 WEEKS OTHER ALMPS (PROGRAMME INTRO TIME)

Months in Locking•in effect Post•programme effect Net effect (weeks) unemployment Estimate Std. Err. Effect, % Estimate Std. Err. Effect, % Estimate Std. Err. 1•6 30•39 •0.41 0.07 •33.5 0.60 0.23 81.6 •1.37 0.29 40•49 •0.14 0.05 •12.9 0.55 0.25 72.8 •1.13 0.50 50•59 •0.15 0.07 •14.2 0.97 0.11 163.9 •7.89 1.38 7•12 30•39 •0.12 0.01 •11.4 0.84 0.18 131.7 •3.33 0.50 40•49 •0.03 0.01 •2.6 0.77 0.12 116.5 •3.16 0.78 50•59 0.04 0.01 3.6 0.67 0.36 96.2 •6.01 2.03 13•18 30•39 •0.20 0.11 •17.8 0.71 0.22 104.4 •2.04 0.78 40•49 •0.13 0.03 •12.0 0.82 0.25 127.6 •2.74 0.76 50•59 •0.25 0.09 •22.0 0.83 0.18 129.4 •5.84 1.60 19•24 30•39 •0.17 0.31 •15.8 0.48 0.20 62.0 •1.06 0.76 40•49 •0.11 0.07 •10.1 0.60 0.23 81.6 •1.12 0.55 50•59 •0.23 0.03 •20.2 0.68 0.21 96.5 •5.49 1.02 >24 30•39 •0.32 0.13 •27.4 0.22 0.17 24.0 0.11 0.05 40•49 •0.09 0.04 •8.9 0.07 0.20 7.6 0.10 0.09 50•59 •0.19 0.28 •17.6 0.55 0.25 73.9 •3.89 1.96

177 SCHOOL OF ECONOMICS AND MANAGEMENT UNIVERSITY OF AARHUS - UNIVERSITETSPARKEN - BUILDING 1322 DK-8000 AARHUS C – TEL. +45 8942 1111 - www.econ.au.dk

PhD Theses:

1999-4 Philipp J.H. Schröder, Aspects of Transition in Central and Eastern Europe.

1999-5 Robert Rene Dogonowski, Aspects of Classical and Contemporary European Fiscal Policy Issues.

1999-6 Peter Raahauge, Dynamic Programming in Computational Economics.

1999-7 Torben Dall Schmidt, Social Insurance, Incentives and Economic Integration.

1999 Jørgen Vig Pedersen, An Asset-Based Explanation of Strategic Advantage.

1999 Bjarke Jensen, Five Essays on Contingent Claim Valuation.

1999 Ken Lamdahl Bechmann, Five Essays on Convertible Bonds and Capital Structure Theory.

1999 Birgitte Holt Andersen, Structural Analysis of the Earth Observation Industry.

2000-1 Jakob Roland Munch, Economic Integration and Industrial Location in Unionized Countries.

2000-2 Christian Møller Dahl, Essays on Nonlinear Econometric Time Series Modelling.

2000-3 Mette C. Deding, Aspects of Income Distributions in a Labour Market Perspective.

2000-4 Michael Jansson, Testing the Null Hypothesis of Cointegration.

2000-5 Svend Jespersen, Aspects of Economic Growth and the Distribution of Wealth.

2001-1 Michael Svarer, Application of Search Models.

2001-2 Morten Berg Jensen, Financial Models for Stocks, Interest Rates, and Options: Theory and Estimation.

2001-3 Niels C. Beier, Propagation of Nominal Shocks in Open Economies.

2001-4 Mette Verner, Causes and Consequences of Interrruptions in the Labour Market.

2001-5 Tobias Nybo Rasmussen, Dynamic Computable General Equilibrium Models: Essays on Environmental Regulation and Economic Growth.

2001-6 Søren Vester Sørensen, Three Essays on the Propagation of Monetary Shocks in Open Economies.

2001-7 Rasmus Højbjerg Jacobsen, Essays on Endogenous Policies under Labor Union Influence and their Implications.

2001-8 Peter Ejler Storgaard, Price Rigidity in Closed and Open Economies: Causes and Effects.

2001 Charlotte Strunk-Hansen, Studies in Financial Econometrics.

2002-1 Mette Rose Skaksen, Multinational Enterprises: Interactions with the Labor Market.

2002-2 Nikolaj Malchow-Møller, Dynamic Behaviour and Agricultural Households in Nicaragua.

2002-3 Boriss Siliverstovs, Multicointegration, Nonlinearity, and Forecasting.

2002-4 Søren Tang Sørensen, Aspects of Sequential Auctions and Industrial Agglomeration.

2002-5 Peter Myhre Lildholdt, Essays on Seasonality, Long Memory, and Volatility.

2002-6 Sean Hove, Three Essays on Mobility and Income Distribution Dynamics.

2002 Hanne Kargaard Thomsen, The Learning organization from a management point of view - Theoretical perspectives and empirical findings in four Danish service organizations.

2002 Johannes Liebach Lüneborg, Technology Acquisition, Structure, and Performance in The Nordic Banking Industry.

2003-1 Carter Bloch, Aspects of Economic Policy in Emerging Markets.

2003-2 Morten Ørregaard Nielsen, Multivariate Fractional Integration and Cointegration.

2003 Michael Knie-Andersen, Customer Relationship Management in the Financial Sector.

2004-1 Lars Stentoft, Least Squares Monte-Carlo and GARCH Methods for American Options.

2004-2 Brian Krogh Graversen, Employment Effects of Active Labour Market Programmes: Do the Programmes Help Welfare Benefit Recipients to Find Jobs?

2004-3 Dmitri Koulikov, Long Memory Models for Volatility and High Frequency Financial Data Econometrics.

2004-4 René Kirkegaard, Essays on Auction Theory.

2004-5 Christian Kjær, Essays on Bargaining and the Formation of Coalitions.

2005-1 Julia Chiriaeva, Credibility of Fixed Exchange Rate Arrangements.

2005-2 Morten Spange, Fiscal Stabilization Policies and Labour Market Rigidities.

2005-3 Bjarne Brendstrup, Essays on the Empirical Analysis of Auctions.

2005-4 Lars Skipper, Essays on Estimation of Causal Relationships in the Danish Labour Market.

2005-5 Ott Toomet, Marginalisation and Discouragement: Regional Aspects and the Impact of Benefits.

2005-6 Marianne Simonsen, Essays on Motherhood and Female Labour Supply.

2005 Hesham Morten Gabr, Strategic Groups: The Ghosts of Yesterday when it comes to Understanding Firm Performance within Industries?

2005 Malene Shin-Jensen, Essays on Term Structure Models, Interest Rate Derivatives and Credit Risk.

2006-1 Peter Sandholt Jensen, Essays on Growth Empirics and Economic Development.

2006-2 Allan Sørensen, Economic Integration, Ageing and Labour Market Outcomes

2006-3 Philipp Festerling, Essays on Competition Policy

2006-4 Carina Sponholtz, Essays on Empirical Corporate Finance

2006-5 Claus Thrane-Jensen, Capital Forms and the Entrepreneur – A contingency approach on new venture creation

2006-6 Thomas Busch, Econometric Modeling of Volatility and Price Behavior in Asset and Derivative Markets

2007-1 Jesper Bagger, Essays on Earnings Dynamics and Job Mobility

2007-2 Niels Stender, Essays on Marketing Engineering

2007-3 Mads Peter Pilkjær Harmsen, Three Essays in Behavioral and Experimental Economics

2007-4 Juanna Schrøter Joensen, Determinants and Consequences of Human Capital Investments

2007-5 Peter Tind Larsen, Essays on Capital Structure and Credit Risk

2008-1 Toke Lilhauge Hjortshøj, Essays on Empirical Corporate Finance – Managerial Incentives, Information Disclosure, and Bond Covenants

2008-2 Jie Zhu, Essays on Econometric Analysis of Price and Volatility Behavior in Asset Markets

2008-3 David Glavind Skovmand, Libor Market Models - Theory and Applications

2008-4 Martin Seneca, Aspects of Household Heterogeneity in New Keynesian Economics

2008-5 Agne Lauzadyte, Active Labour Market Policies and Labour Market Transitions in Denmark: an Analysis of Event History Data